[go: up one dir, main page]

CN109682976B - Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion - Google Patents

Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion Download PDF

Info

Publication number
CN109682976B
CN109682976B CN201910153512.4A CN201910153512A CN109682976B CN 109682976 B CN109682976 B CN 109682976B CN 201910153512 A CN201910153512 A CN 201910153512A CN 109682976 B CN109682976 B CN 109682976B
Authority
CN
China
Prior art keywords
model
moment
current
value
blood glucose
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.)
Active
Application number
CN201910153512.4A
Other languages
Chinese (zh)
Other versions
CN109682976A (en
Inventor
于霞
崔悦
刘建昌
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.)
Northeastern University China
Original Assignee
Northeastern University China
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 Northeastern University China filed Critical Northeastern University China
Priority to CN201910153512.4A priority Critical patent/CN109682976B/en
Publication of CN109682976A publication Critical patent/CN109682976A/en
Application granted granted Critical
Publication of CN109682976B publication Critical patent/CN109682976B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/48—Biological material, e.g. blood, urine; Haemocytometers
    • G01N33/50—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing
    • G01N33/66—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing involving blood sugars, e.g. galactose

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Hematology (AREA)
  • Molecular Biology (AREA)
  • Urology & Nephrology (AREA)
  • Biomedical Technology (AREA)
  • Chemical & Material Sciences (AREA)
  • Immunology (AREA)
  • Food Science & Technology (AREA)
  • Analytical Chemistry (AREA)
  • Microbiology (AREA)
  • Cell Biology (AREA)
  • Biotechnology (AREA)
  • Medicinal Chemistry (AREA)
  • Physics & Mathematics (AREA)
  • Diabetes (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Pathology (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
  • Investigating Or Analysing Biological Materials (AREA)

Abstract

本发明涉及一种基于多模型融合的连续血糖监测传感器在线故障检测方法，包括如下步骤：S1、获取在线CGM监测信号数据；S2、将获取的在线CGM监测信号数据输入多模型融合算法模型中，获取在线预测误差；S3、将获取的在线预测误差和历史预测误差结合计算获得在线时刻的熵值；S4、将计算获得的在线时刻的熵值Ji1、Ji2分别与当前时刻的阈值Tkl1、Tkl2比较；若当前时刻的熵值Ji1、Ji2不全大于当前时刻的阈值Tkl1、Tkl2，则判断当前血糖监测传感器工作正常；若当前时刻的熵值Ji1、Ji2均大于当前时刻的阈值Tkl1、Tkl2，则判断当前血糖监测传感器工作异常。本发明提供的检测方法具有检测精度高的优点。

Figure 201910153512

The invention relates to an online fault detection method for a continuous blood glucose monitoring sensor based on multi-model fusion, comprising the following steps: S1, acquiring online CGM monitoring signal data; S2, inputting the acquired online CGM monitoring signal data into a multi-model fusion algorithm model, Obtain the online prediction error; S3, combine the obtained online prediction error and the historical prediction error to obtain the entropy value of the online moment; S4, respectively calculate the entropy values J i1 and J i2 of the online moment obtained by the calculation and the threshold value T kl1 at the current moment and T kl2 are compared; if the entropy values J i1 and J i2 at the current moment are not all greater than the threshold values T kl1 and T kl2 at the current moment, it is judged that the current blood glucose monitoring sensor is working normally; if the entropy values J i1 and J i2 at the current moment are both greater than The thresholds T kl1 and T kl2 at the current moment determine that the current blood glucose monitoring sensor is abnormally working. The detection method provided by the present invention has the advantages of high detection accuracy.

Figure 201910153512

Description

Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion
Technical Field
The invention belongs to the technical field of blood glucose monitoring, and particularly relates to an online fault detection method of a continuous blood glucose monitoring sensor based on multi-model fusion.
Background
The Artificial Pancreas (AP) system provides automatic regulation of Blood Glucose Concentration (BGC) for type 1 diabetes (T1D) patients, and it is composed of three major components: a continuous blood glucose monitoring (CGM) sensor, a controller that calculates an insulin infusion rate based on the CGM signal, and an insulin pump that delivers the amount of insulin calculated by the controller to the patient. The patient with T1D can know the fluctuation of blood sugar more comprehensively by continuously monitoring the blood sugar, thereby realizing better blood sugar control. However, in real life, the measurement result of the continuous blood glucose monitoring sensor is influenced by various factors, so that the measurement result is inaccurate, and the artificial pancreas control system infuses wrong amount of insulin according to wrong measurement value, which finally causes the patient to have hyperglycemia and even threaten life in severe cases.
Methods have been proposed for detecting incorrect measurements of continuous glucose monitoring sensors, which are mainly classified into two categories, one being model-based methods that do not require large amounts of historical data, and that determine whether a system has failed by merely modeling blood glucose data and comparing the predicted values of the model with the measured values; while another class is based on data-driven methods that strongly depend on the size and performance of the data set, require large amounts of historical data, and calculate their confidence limits from statistical analysis, a representative of such methods is the PCA method. The currently common modeling methods include an autoregressive moving average method, a Support Vector Machine (SVM), Kalman Filtering (KF), a Gaussian Mixture Model (GMM), Recursive Least Squares (RLS), a model based on a kernel filtering algorithm and the like, most of the methods assume that data meet Gaussian distribution and only consider the influence brought by error information at the current moment, and cannot effectively distinguish the rapid change of blood glucose data and the abnormity of sensor signals.
Disclosure of Invention
Technical problem to be solved
Aiming at the existing technical problems, the invention provides a continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion, which solves the problems of low accuracy of detection results and the like in the prior art.
(II) technical scheme
In order to achieve the purpose, the invention adopts the main technical scheme that:
a continuous blood sugar monitoring sensor online fault detection method based on multi-model fusion comprises the following steps:
s1, acquiring on-line CGM monitoring signal data;
s2, inputting the acquired online CGM monitoring signal data into a multi-model fusion algorithm model to acquire online prediction errors;
s3, combining the obtained online prediction error and the historical prediction error to calculate and obtain an entropy value of the online time;
s4, calculating the obtained entropy value J of the online timei1、Ji2Respectively with the threshold value T of the current momentkl1、Tkl2Comparing;
if the entropy value J of the current timei1、Ji2Not all greater than the threshold value T at the present momentkl1、Tkl2Judging that the current blood sugar monitoring sensor works normally;
if the entropy value J of the current timei1、Ji2Are all larger than the threshold value T of the current momentkl1、Tkl2And judging that the current blood sugar monitoring sensor works abnormally.
Preferably, when the current blood glucose monitoring sensor is judged to work normally, the method further comprises the following steps:
S40A1, secondary detection;
S40A2, return to step S1.
Preferably, when the current blood glucose monitoring sensor is judged to work abnormally, the method further comprises the following steps:
S40B1, returning to the step S1;
and S40B2, replacing the monitoring signal data in the step 1 with the predicted value of the current multi-model fusion algorithm model.
Preferably, the step S40a1 further includes:
a1, calculating two entropy values J of the current timei1And Ji2；
A2, two entropy values Ji1、Ji2Respectively with a threshold value Tkl1、Tkl2Comparing;
if entropy value Ji1And Ji2All exceed the threshold value Tkl1、Tkl2；
Judging that the current blood sugar monitoring sensor works abnormally, replacing a measured value with a predicted value of the model, and reconstructing a model prediction error at the current moment;
otherwise, judging that the current blood sugar monitoring sensor works normally, and updating the parameters of the model by using the measured values.
Preferably, after the step a2, determining that the current blood glucose monitoring sensor is abnormal in operation, the method further includes the following steps:
b1, calculating entropy J of current time1And J2；
B2, judging the entropy J of the current time1And J2If so, execute B3, otherwise execute B4;
b3, judging whether the measured value at the previous moment is a fault value, if not, setting a threshold value as a 3 delta confidence interval of the entropy value in the window, and storing the entropy value at the current moment and the entropy value at the previous moment, otherwise, updating and storing the entropy value at the current moment;
and B4, judging whether the measured value at the previous moment is a fault value, if so, setting a threshold value to be 95% of the maximum change value when the entropy value at the latest moment is reduced, otherwise, setting the threshold value to be reasonable.
Preferably, the entropy J of the current time is calculated in the step B11And J2Also comprises the following steps:
c1, obtaining a model prediction error of the multi-model fusion algorithm model at the current moment;
c2, obtaining model prediction errors of historical time and recent time;
c3, calculating the mean and variance of the prediction errors of the historical time model and the recent time model as p (x) and p1(x) Mean and variance of the distribution;
c4, calculating the mean and variance of the prediction errors of the model including the historical time and the latest time after the prediction error of the model at the current time as q (x) and q1(x) Mean and variance of the distribution;
c5, calculating by using calculation formula of KL divergenceEntropy of previous time J1And J2；
Preferably, the calculation formula of the KL divergence in step C5 is:
Figure BDA0001982162970000041
wherein p (x) and q (x) are two univariate normal distributions and satisfy p to N (mu)0，σ0) And q to N (mu)1，σ1)。
Preferably, the multi-model fusion algorithm model comprises the following steps:
d1, acquiring blood glucose data g (t) measured by a continuous blood glucose monitoring sensor;
d2, reconstructing the acquired data by using a sliding window with length L to obtain the following input matrix and output matrix:
Figure BDA0001982162970000042
y (N × 1) bis [ g (L + PH) g (L +1+ PH) ]. g (k)]T(2)
Wherein x (i) ═ g (i) g (i +1) … g (i + L-1) ], N ═ K-PH-L +1 is the number of samples to predict, K is the number of samples of the original time series, PH denotes the number of steps to predict ahead, and i denotes the current time;
d3, respectively modeling the reconstructed data by using SVM and RLS algorithm to generate a predicted value y of the modelSVMAnd yRLS；
D4, calculating the average value of the historical prediction errors of each model;
d5, judging whether the model prediction error of each model at the previous moment is more than 3, if so, executing a step D6, otherwise, executing a step D7;
d6, comparing the prediction errors mean of the two modelssvmAnd meanrlsIf mean, ofsvm＜meanrlsAnd the final predicted value of the model is Y ═ YSVMOtherwise Y ═ YRLS’And then the process is ended;
d7, calculating the weight of each model according to the prediction error of each model;
d8, calculating an expression of the model predicted value after multi-model fusion as follows:
Y＝ySVM×wsvm+yRLS×wrls
wherein, wsvm+vrls＝1。
Preferably, the calculation formula for calculating the mean value of the historical prediction errors of each model in the step D4 is as follows:
Figure BDA0001982162970000051
Figure BDA0001982162970000052
wherein, errorsvmAnd errorrlsRespectively predicting errors of the SVM and the RLS model, wherein n is the number of model predicting errors and satisfies that i is larger than n;
preferably, the formula for calculating the weights of the models in step D7 is as follows:
Figure BDA0001982162970000053
Figure BDA0001982162970000054
wherein, wsvmThe weight occupied by the SVM model; w is arlsIs the weight taken up by the RLS model.
(III) advantageous effects
The invention has the beneficial effects that: according to the online fault detection method of the continuous blood glucose monitoring sensor based on multi-model fusion, the model prediction value of the current moment is generated through the multi-model fusion method according to the collected historical data, the prediction error of the current moment is analyzed by using the historical error of model prediction, the dynamic threshold updating strategy is designed by considering the dynamics of blood glucose data, and the rapid change of the blood glucose data and the abnormity of the sensor signal can be effectively distinguished. In addition, the invention can process the fault signal, can effectively avoid the insulin pump from injecting wrong amount of insulin according to wrong blood glucose data information, and further reduce the influence on the life safety of the patient.
The multi-model fusion prediction method can improve the prediction capability of the model, reduce the influence of the prediction error of the model on the result, and take the dynamics of blood glucose data into consideration. The support vector machine method can better fit the nonlinear data, but has larger prediction error near an extreme point with larger fluctuation, so the prediction precision of the model is improved by adopting a multi-model prediction method which integrates the two methods based on the analysis.
Drawings
FIG. 1 is a schematic flow chart of a continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion according to the present invention;
FIG. 2 is a schematic flow chart of a multi-model fusion algorithm model in the online fault detection method for a continuous blood glucose monitoring sensor based on multi-model fusion according to the present invention;
FIG. 3 is a schematic flow chart of dynamic threshold updating in a continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion according to the present invention;
FIG. 4 is a schematic diagram of an algorithm flow of a continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion according to the present invention;
FIG. 5 is a schematic diagram of an algorithm flow of a multi-model fusion algorithm model in the online fault detection method for a continuous blood glucose monitoring sensor based on multi-model fusion according to the present invention;
FIG. 6 is a schematic diagram showing comparison results of different models for blood glucose data prediction in a continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion according to the present invention;
FIG. 7 is a schematic diagram of KL divergence-based online fault detection in a continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion according to the present invention;
FIG. 8 is a schematic diagram of a calculation flow of KL divergence in the online fault detection method of a continuous blood glucose monitoring sensor based on multi-model fusion provided by the present invention.
Detailed Description
For the purpose of better explaining the present invention and to facilitate understanding, the present invention will be described in detail by way of specific embodiments with reference to the accompanying drawings.
As shown in fig. 1 and 4: the embodiment discloses a continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion, which comprises the following steps:
s1, acquiring on-line CGM monitoring signal data;
s2, inputting the acquired online CGM monitoring signal data into a multi-model fusion algorithm model to acquire online prediction errors;
s3, combining the obtained online prediction error and the historical prediction error to calculate and obtain an entropy value of the online time;
s4, calculating the obtained entropy value J of the online timei1、Ji2Respectively with the threshold value T of the current momentkl1、Tkl2Comparing;
if the entropy value J of the current timei1、Ji2Not all greater than the threshold value T at the present momentkl1、Tkl2Judging that the current blood sugar monitoring sensor works normally;
if the entropy value J of the current timei1、Ji2Are all larger than the threshold value T of the current momentkl1、Tkl2And judging that the current blood sugar monitoring sensor works abnormally.
In this embodiment, when it is determined that the current blood glucose monitoring sensor is working normally, the method further includes:
S40A1, secondary detection;
S40A2, return to step S1.
In this embodiment, when it is determined that the current blood glucose monitoring sensor is abnormal, the method further includes:
S40B1, returning to the step S1;
and S40B2, replacing the monitoring signal data in the step 1 with the predicted value of the current multi-model fusion algorithm model.
In this embodiment, the step S40a1 further includes:
a1, calculating two entropy values J of the current timei1And Ji2；
A2, two entropy values Ji1、Ji2Respectively with a threshold value Tkl1、Tkl2Comparing;
if entropy value Ji1And Ji2All exceed the threshold value Tkl1、Tkl2；
Judging that the current blood sugar monitoring sensor works abnormally, replacing a measured value with a predicted value of the model, and reconstructing a model prediction error at the current moment;
otherwise, judging that the current blood sugar monitoring sensor works normally, and updating the parameters of the model by using the measured values.
The secondary detection can more comprehensively check whether the work of the sensor is abnormal or not, and the accuracy of the detection result is greatly improved.
As shown in fig. 3: in this embodiment, after the step a2 of determining that the current blood glucose monitoring sensor is abnormal, the method further includes the following steps:
b1, calculating entropy J of current time1And J2；
B2, judging the entropy J of the current time1And J2If so, execute B3, otherwise execute B4;
b3, judging whether the measured value at the previous moment is a fault value, if not, setting a threshold value as a 3 delta confidence interval of the entropy value in the window, and storing the entropy value at the current moment and the entropy value at the previous moment, otherwise, updating and storing the entropy value at the current moment;
and B4, judging whether the measured value at the previous moment is a fault value, if so, setting a threshold value to be 95% of the maximum change value when the entropy value at the latest moment is reduced, otherwise, setting the threshold value to be reasonable.
The dynamic threshold updating strategy provided by the method can effectively distinguish the rapid change of the blood glucose data from the abnormity of the sensor signal.
As shown in fig. 8: in the embodiment, the entropy J of the current time is calculated in the step B11And J2Also comprises the following steps:
c1, obtaining a model prediction error of the multi-model fusion algorithm model at the current moment;
c2, obtaining model prediction errors of historical time and recent time;
c3, calculating the mean and variance of the prediction errors of the historical time model and the recent time model as p (x) and p1(x) Mean and variance of the distribution;
c4, calculating the mean and variance of the prediction errors of the model including the historical time and the latest time after the prediction error of the model at the current time as q (x) and q1(x) Mean and variance of the distribution;
c5, calculating entropy J of current time by using calculation formula of KL divergence1And J2；
The calculation formula of the KL divergence in step C5 in this embodiment is as follows:
Figure BDA0001982162970000081
wherein p (x) and q (x) are two univariate normal distributions and satisfy p to N (mu)0，σ0) And q to N (mu)1，σ1)。
As shown in fig. 2 and 5: the multi-model fusion algorithm model in the embodiment includes the following steps:
d1, acquiring blood glucose data g (t) measured by a continuous blood glucose monitoring sensor;
d2, reconstructing the acquired data by using a sliding window with length L to obtain the following input matrix and output matrix:
Figure BDA0001982162970000091
y (N × 1) bis [ g (L + PH) g (L +1+ PH) ]. g (k)]T(2)
Wherein x (i) ═ g (i) g (i +1) … g (i + L-1) ], N ═ K-PH-L +1 is the number of samples to predict, K is the number of samples of the original time series, PH denotes the number of steps to predict ahead, and i denotes the current time;
d3, respectively modeling the reconstructed data by using SVM and RLS algorithm to generate a predicted value y of the modelSVMAnd yRLS；
D4, calculating the average value of the historical prediction errors of each model;
d5, judging whether the model prediction error of each model at the previous moment is more than 3, if so, executing a step D6, otherwise, executing a step D7;
d6, comparing the prediction errors mean of the two modelssvmAnd meanrlsIf mean, ofsvm＜meanrlsAnd the final predicted value of the model is Y ═ YSVMOtherwise Y ═ YRLS’And then the process is ended;
d7, calculating the weight of each model according to the prediction error of each model;
d8, calculating an expression of the model predicted value after multi-model fusion as follows:
Y＝ySVM×wsvm+yRLS×wrls
wherein, wsvm+wrls＝1。
In the embodiment, the multi-model fusion algorithm model generates the model prediction value at the current moment through a multi-model fusion method according to the collected historical data, and analyzes the prediction error at the current moment by using the historical error of the model prediction, so that the detection result is more accurate.
It should be noted that: the calculation formula for calculating the mean value of the historical prediction errors of each model in the step D4 is as follows:
Figure BDA0001982162970000101
Figure BDA0001982162970000102
wherein, errorsvmAnd errorrlsRespectively predicting errors of the SVM and the RLS model, wherein n is the number of model predicting errors and satisfies that i is larger than n;
it should be noted that: the formula for calculating the weights of the models in step D7 is as follows:
Figure BDA0001982162970000103
Figure BDA0001982162970000104
wherein, wsvmThe weight occupied by the SVM model; w is arlsIs the weight taken up by the RLS model.
1. Performance verification of multi-model fusion prediction method
The method presented in this example was validated on a type I diabetes metabolic simulator at the university of feignia/padova italy. Experimental data were obtained by sampling at five minute intervals, the data set contained 6-day blood glucose data for three types of patients, adult, juvenile and child, and the performance of each model was evaluated by using a one-step prediction method, with the results shown in fig. 6. The comparison result of the model prediction performance by using the single model and the multi-model fusion method is shown in the table 1, and the fitting degree between the multi-model fusion prediction method and the blood glucose data is better compared with the single model prediction method by analyzing the prediction results of the models on the blood glucose data of the three types of patients, so that the rationality of the algorithm provided by the invention is illustrated.
TABLE 1 comparison of prediction Performance of Single model and Multi-model hybrid prediction methods
Figure BDA0001982162970000105
Figure BDA0001982162970000111
2. Performance verification of online fault detection algorithms
The continuous blood glucose monitoring sensor has six common fault types, namely spike, drift, step, pressure sensing sensor attenuation, signal loss and stagnation. Wherein the loss and stagnation of signals are easy to be detected, the other four fault signals are randomly added to the normal blood glucose data in a period of 4 hours in the simulation process, the amplitude of the fault signal is set to be 10%, and fig. 7 is a fault detection result of an online fault detection algorithm on the blood glucose data of children. As can be seen from the figure, the detection method provided by the invention can effectively detect small fault signals of four types of faults in time and reconstruct the fault signals so as to reduce the damage caused when the continuous blood sugar monitoring sensor fails and effectively improve the performance of the continuous blood sugar monitoring sensor.
The technical principles of the present invention have been described above in connection with specific embodiments, which are intended to explain the principles of the present invention and should not be construed as limiting the scope of the present invention in any way. Based on the explanations herein, those skilled in the art will be able to conceive of other embodiments of the present invention without inventive efforts, which shall fall within the scope of the present invention.

Claims (5)

1.一种基于多模型融合的连续血糖监测传感器在线故障检测方法，其特征在于，包括如下步骤：1. a continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion, is characterized in that, comprises the steps: S1、获取在线CGM监测信号数据；S1. Obtain online CGM monitoring signal data; S2、将获取的在线CGM监测信号数据输入基于递归最小二乘和支持向量机的多模型融合算法模型中，获取在线预测误差；S2. Input the obtained online CGM monitoring signal data into a multi-model fusion algorithm model based on recursive least squares and support vector machines to obtain online prediction errors; S3、将获取的在线预测误差和历史预测误差结合计算获得在线时刻的熵值；S3, combining the obtained online prediction error and historical prediction error to obtain the entropy value at the online moment; S4、将计算获得的在线时刻的熵值Ji1、Ji2分别与当前时刻的阈值Tkl1、Tkl2比较；S4, compare the entropy values J i1 and J i2 of the online moment obtained by calculation with the threshold values T kl1 and T kl2 at the current moment, respectively; 若当前时刻的熵值Ji1、Ji2不全大于当前时刻的阈值Tkl1、Tkl2，则判断当前血糖监测传感器工作正常；If the entropy values J i1 and J i2 at the current moment are not all greater than the thresholds T kl1 and T kl2 at the current moment, it is determined that the current blood glucose monitoring sensor is working normally; 若当前时刻的熵值Ji1、Ji2均大于当前时刻的阈值Tkl1、Tkl2，则判断当前血糖监测传感器工作异常；If the entropy values J i1 and J i2 at the current moment are both greater than the thresholds T kl1 and T kl2 at the current moment, it is determined that the current blood glucose monitoring sensor is working abnormally; 在判断当前血糖监测传感器工作正常时，所述方法还包括：When judging that the current blood glucose monitoring sensor works normally, the method further includes: S40A1、二次检测；S40A1, secondary detection; S40A2、返回步骤S1；S40A2, returning to step S1; 所述基于递归最小二乘和支持向量机的多模型融合算法模型包括如下步骤：The multi-model fusion algorithm model based on recursive least squares and support vector machine includes the following steps: D1、获取连续血糖监测传感器测量到的血糖数据g(t)；D1. Obtain the blood glucose data g(t) measured by the continuous blood glucose monitoring sensor; D2、使用一个长度为L的滑动窗口对获取的数据进行重构，得到以下形式的输入矩阵和输出矩阵：D2. Use a sliding window of length L to reconstruct the acquired data to obtain the input matrix and output matrix of the following form:
Figure FDA0002429393810000011
Figure FDA0002429393810000011
 Y(N*1)＝[g(L+PH) g(L+1+PH) ... g(K)]T (2)Y(N*1)=[g(L+PH) g(L+1+PH) ... g(K)] T (2) 其中，x(i)＝[g(i) g(i+1)…g(i+L-1)]，N＝K-PH-L+1是用来预测的样本数，K是原始时间序列的样本数，PH表示超前预测的步数，i表示当前时刻；Where, x(i)=[g(i) g(i+1)...g(i+L-1)], N=K-PH-L+1 is the number of samples used for prediction, K is the original time The number of samples in the sequence, PH represents the number of steps ahead of the prediction, and i represents the current moment; D3、使用SVM和RLS算法分别对重构后的数据进行建模，产生模型的预测值ySVM和yRLS；D3, use the SVM and RLS algorithms to model the reconstructed data respectively, and generate the predicted values y SVM and y RLS of the model; D4、计算各模型历史预测误差的均值；D4. Calculate the mean value of historical prediction errors of each model; D5、判断各模型上一时刻的模型预测误差是否大于3，如果是，执行步骤D6，否则执行步骤D7；D5, determine whether the model prediction error of each model at the last moment is greater than 3, if so, go to step D6, otherwise go to step D7; D6、比较两个模型预测误差meansvm：和meanrls的大小，如果meansvm＜meanrls，则模型的最终预测值为Y＝ySVM，否则Y＝yRLS，并结束；D6. Compare the prediction error mean svm of the two models: and the size of mean rls , if mean svm <mean rls , the final prediction value of the model is Y=y SVM , otherwise Y=y RLS , and end; D7、根据各模型的预测误差，计算各模型的权重；D7. Calculate the weight of each model according to the prediction error of each model; D8、计算多模型融合后的模型预测值的表达式如下式：D8. The expression for calculating the model prediction value after multi-model fusion is as follows: Y＝ySVM×wsvm+yRLS×wrls Y=y SVM ×w svm +y RLS ×w rls 其中，wsvm+wrls＝1。where w svm + w rls =1. 2.根据权利要求1所述的检测方法，其特征在于，在判断当前血糖监测传感器工作异常时，所述方法还包括：2. The detection method according to claim 1, wherein when judging that the current blood glucose monitoring sensor is abnormally working, the method further comprises: S40B1、返回步骤S1；S40B1, returning to step S1; S40B2、将当前的多模型融合算法模型的预测值代替步骤1中的监测信号数据。S40B2: Replace the monitoring signal data in step 1 with the predicted value of the current multi-model fusion algorithm model. 3.根据权利要求1所述的检测方法，其特征在于，所述步骤S40A1还包括：3. The detection method according to claim 1, wherein the step S40A1 further comprises: A1、计算当前时刻的两个熵值Ji1和Ji2；A1. Calculate two entropy values J i1 and J i2 at the current moment; A2、将两个熵值Ji1、Ji2分别与阈值Tkl1、Tkl2进行比较；A2. Compare the two entropy values J i1 and J i2 with the thresholds T kl1 and T kl2 respectively; 若熵值Ji1和Ji2均超过阈值Tkl1、Tkl2；If the entropy values J i1 and J i2 both exceed the thresholds T kl1 and T kl2 ; 则判断当前血糖监测传感器工作异常，用模型的预测值代替测量值，并对当前时刻的模型预测误差进行重构；Then it is judged that the current blood glucose monitoring sensor is abnormal, the predicted value of the model is used to replace the measured value, and the model prediction error at the current moment is reconstructed; 否则判断当前血糖监测传感器工作正常，利用测量值对模型的参数进行更新。Otherwise, it is judged that the current blood glucose monitoring sensor is working normally, and the parameters of the model are updated by using the measured value. 4.根据权利要求3所述的检测方法，其特征在于，所述步骤A2中判断当前血糖监测传感器工作异常后还包括如下步骤：4. detection method according to claim 3 is characterized in that, in described step A2, after judging that current blood glucose monitoring sensor works abnormally, also comprises the following steps: B1、计算当前时刻的熵值J1和J2；B1. Calculate the entropy values J 1 and J 2 at the current moment; B2、判断当前时刻的熵值J1和J2的值是否上升，若上升，则执行B3，否则执行B4；B2. Determine whether the values of the entropy values J 1 and J 2 at the current moment rise, if they rise, execute B3, otherwise execute B4; B3、判断上一时刻测量值是否为故障值，若否，则设定阈值为窗口内熵值的3δ置信区间，并存储当前时刻的熵值和上一时刻的熵值，否则更新存储当前时刻的熵值；B3. Determine whether the measured value at the last moment is a fault value, if not, set the threshold as the 3δ confidence interval of the entropy value in the window, and store the entropy value at the current moment and the entropy value at the previous moment, otherwise update and store the current moment the entropy value; B4、判断上一时刻测量值是否为故障值，若是，则设定阈值为最近时刻熵值下降时最大变化值的95％，否则阈值合理。B4. Determine whether the measured value at the last moment is a fault value, if so, set the threshold to 95% of the maximum change value when the entropy value drops at the latest moment, otherwise the threshold is reasonable. 5.根据权利要求4所述的检测方法，其特征在于，所述步骤B1中计算当前时刻的熵值J1和J2还包括如下步骤：5. detection method according to claim 4 is characterized in that, in described step B1, calculating the entropy value J 1 and J 2 of current moment also comprises the following steps: C1、获取当前时刻的多模型融合算法模型的模型预测误差；C1. Obtain the model prediction error of the multi-model fusion algorithm model at the current moment; C2、获取历史时刻和最近时刻的模型预测误差；C2. Obtain the model prediction error at the historical moment and the most recent moment; C3、分别计算历史时刻和最近时刻模型预测误差的均值和方差，作为p(x)和p1(x)分布的均值和方差；C3. Calculate the mean and variance of the model prediction errors at the historical moment and the most recent moment, respectively, as the mean and variance of the p(x) and p 1 (x) distributions; C4、分别计算包含当前时刻模型预测误差后的历史时刻和最近时刻模型预测误差的均值和方差，作为q(x)和q1(x)分布的均值和方差；C4. Calculate the mean and variance of the model prediction error at the historical moment including the model prediction error at the current moment and at the latest moment, respectively, as the mean and variance of the q(x) and q 1 (x) distributions; C5、采用KL散度的计算公式，计算当前时刻的熵值J1和J2。C5. Using the calculation formula of KL divergence, calculate the entropy values J 1 and J 2 at the current moment.
CN201910153512.4A 2019-02-28 2019-02-28 Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion Active CN109682976B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910153512.4A CN109682976B (en) 2019-02-28 2019-02-28 Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910153512.4A CN109682976B (en) 2019-02-28 2019-02-28 Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion

Publications (2)

Publication Number Publication Date
CN109682976A CN109682976A (en) 2019-04-26
CN109682976B true CN109682976B (en) 2020-05-22

Family

ID=66196311

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910153512.4A Active CN109682976B (en) 2019-02-28 2019-02-28 Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion

Country Status (1)

Country Link
CN (1) CN109682976B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA3146849A1 (en) * 2019-07-10 2021-01-14 University Of Virginia Patent Foundation System and method for online domain adaptation of models for hypoglycemia prediction in type 1 diabetes
CN110448306B (en) * 2019-07-30 2023-02-07 东北大学 Online fault detection and diagnosis method based on continuous blood glucose monitoring system
CN110379503B (en) * 2019-07-30 2023-08-25 东北大学 Online fault detection and diagnosis system based on continuous blood glucose monitoring system
CN110490266B (en) * 2019-08-23 2022-04-22 北京邮电大学 Sensor data uploading and sensor fault detection method and device
JP7382791B2 (en) * 2019-10-30 2023-11-17 株式会社日立製作所 Abnormality determination device, vehicle support system
CN111991003A (en) * 2020-08-12 2020-11-27 上海萌草科技有限公司 Savitzky-Golay filtering-based continuous blood glucose smoothing method, device, equipment and storage medium
CN112285413B (en) * 2020-10-15 2024-06-14 中国第一汽车股份有限公司 Current value determining method, device, controller, medium and battery management system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103344881A (en) * 2013-06-27 2013-10-09 黑龙江科技大学 Grid fault diagnosing method based on combined type cross entropy algorithm
CN106094749A (en) * 2016-06-17 2016-11-09 浙江大学 Based on the nonlinear fault detection method and the application that improve nuclear entropy component analysis
CN107137093A (en) * 2017-04-20 2017-09-08 浙江大学 A kind of continuous blood sugar monitoring device for including abnormal plasma glucose probability alarm
CN109378072A (en) * 2018-10-13 2019-02-22 中山大学 An early warning method for abnormal fasting blood glucose value based on ensemble learning fusion model

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7436511B2 (en) * 1999-01-22 2008-10-14 Sensys Medical, Inc. Analyte filter method and apparatus

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103344881A (en) * 2013-06-27 2013-10-09 黑龙江科技大学 Grid fault diagnosing method based on combined type cross entropy algorithm
CN106094749A (en) * 2016-06-17 2016-11-09 浙江大学 Based on the nonlinear fault detection method and the application that improve nuclear entropy component analysis
CN107137093A (en) * 2017-04-20 2017-09-08 浙江大学 A kind of continuous blood sugar monitoring device for including abnormal plasma glucose probability alarm
CN109378072A (en) * 2018-10-13 2019-02-22 中山大学 An early warning method for abnormal fasting blood glucose value based on ensemble learning fusion model

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Fault Detection in Continuous Glucose Monitoring Sensors for Artifical Pancreas Systems;Xia Yu，et al.;《IFAC PapersOnLine》;20181231;第714-719页 *
Hybrid Online Sensor Error Detection and Functional Redundancy for Artificial Pancreas Control Systems;Jianyuan Feng，et al.;《11th IFAC Symposium on Dynamics and Control of Process Systems,including Biosystems》;20161231;第753-758页 *
Model-Fusion-Based Online Glucose Concentration Predictions in People with Type 1 Diabetes;Xia Yu，et al.;《Control Eng Pract》;20190201;第71卷;第129-141页 *
Monitoring and Fault Detection of Continuous Glucose Sensor Measurements;Kamuran Turksoy，et al.;《American Control Conference》;20151231;第5091-6096页 *

Also Published As

Publication number Publication date
CN109682976A (en) 2019-04-26

Similar Documents

Publication Publication Date Title
CN109682976B (en) Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion
EP2727530B1 (en) Calibration of in vivo sensors used to measure analyte concentration
US20160223514A1 (en) Method for denoising and data fusion of biophysiological rate features into a single rate estimate
Meneghetti et al. Model-based detection and classification of insulin pump faults and missed meal announcements in artificial pancreas systems for type 1 diabetes therapy
RU2018134296A (en) PROBABILITY-BASED CONTROLLER REINFORCEMENT
CA3066900C (en) Method and state machine system for detecting an operation status for a sensor
Zhao et al. Rapid model identification for online subcutaneous glucose concentration prediction for new subjects with type I diabetes
CN111326255A (en) Blood glucose concentration correction method based on big data, storage medium and system
Chen et al. Committed moving horizon estimation for meal detection and estimation in type 1 diabetes
CN115482932A (en) Multivariate blood glucose prediction algorithm based on transfer learning and glycosylated hemoglobin
CN110448306B (en) Online fault detection and diagnosis method based on continuous blood glucose monitoring system
US11571169B2 (en) Probability-based detector and controller apparatus, method, computer program
CN117766145A (en) Continuous blood glucose prediction model construction method, blood glucose prediction method and device
CN110379503B (en) Online fault detection and diagnosis system based on continuous blood glucose monitoring system
Zheng et al. Unannounced meal detection for artificial pancreas systems using extended isolation forest
CN109567869A (en) Accelerate movable method and system on a kind of processing fetal heart rate curve
CN119153070A (en) Multi-parameter monitoring data analysis method
CN109712715A (en) A kind of scoring processing method and processing device of physiological data
CN112370013A (en) Method and system for determining sleep stage
WO2020220403A1 (en) Fatigue state detection method, apparatus and device, and storage medium
CN110841152A (en) Heatable pressure device for blood transfusion of hematology department
Manzoni et al. Detection of glucose sensor faults in an artificial pancreas via whiteness test on Kalman filter residuals
Staal et al. Meal estimation from continuous glucose monitor data using kalman filtering and hypothesis testing
Song et al. A classification-based fault detection method for Continuous glucose monitoring (CGM)
Yu et al. Rapid model identification for online glucose prediction of new subjects with type 1 diabetes using model migration method

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