CN109682976A - Continuous blood sugar based on multi-model fusion monitors the online fault detection method of sensor - Google Patents
Continuous blood sugar based on multi-model fusion monitors the online fault detection method of sensor Download PDFInfo
- Publication number
- CN109682976A CN109682976A CN201910153512.4A CN201910153512A CN109682976A CN 109682976 A CN109682976 A CN 109682976A CN 201910153512 A CN201910153512 A CN 201910153512A CN 109682976 A CN109682976 A CN 109682976A
- Authority
- CN
- China
- Prior art keywords
- model
- entropy
- current time
- value
- svm
- 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
Links
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
Abstract
The present invention relates to a kind of continuous blood sugars based on multi-model fusion to monitor the online fault detection method of sensor, includes the following steps: S1, obtains online CGM monitoring signals data；In S2, the online CGM monitoring signals data that will acquire input multi-model blending algorithm model, on-line prediction error is obtained；S3, the on-line prediction error that will acquire and historical forecast error combine the entropy for calculating and obtaining the online moment；S4, the entropy J that the online moment obtained will be calculatedi1、Ji2Respectively with the threshold value T at current timekl1、Tkl2Compare；If the entropy J at current timei1、Ji2It is not greater than the threshold value T at current time entirelykl1、Tkl2, then judge that current blood glucose monitoring working sensor is normal；If the entropy J at current timei1、Ji2It is all larger than the threshold value T at current timekl1、Tkl2, then judge that current blood glucose monitoring working sensor is abnormal.Detection method provided by the invention has the advantages that detection accuracy is high.
Description
Technical field
The invention belongs to glucose monitoring techniques fields more particularly to a kind of continuous blood sugar based on multi-model fusion to monitor and pass
The online fault detection method of sensor.
Background technique
Artificial pancreas (AP) system provides the automatic adjustment of blood sugar concentration (BGC) for type 1 diabetes (T1D) patient, it leads
To be made of three parts: continuous blood sugar monitors (CGM) sensor, and the control of infusion of insulin rate is calculated based on CGM signal
Device, and the amount of insulin that controller calculates is passed into the insulin pump of patient.Patient with T1D is by carrying out blood glucose
Continuous monitoring, can understand the fluctuation situation of blood glucose, so that the control to blood glucose be better achieved more fully hereinafter.However,
The measurement result of continuous blood sugar monitoring sensor is influenced by many factors in real life, and causes measurement result inaccurate,
Artificial pancreas control system eventually causes patient and height blood occurs according to the insulin of the measured value infusion number of errors of mistake
Sugared phenomenon, when serious even threat to life.
Certain methods have been proposed for the incorrect measurement of detection continuous blood sugar monitoring sensor, these methods at present
Two classes are broadly divided into, one kind is the method based on model, it does not need a large amount of historical data, only by establishing to blood glucose level data
Model simultaneously is compared to the predicted value of model and measured value to judge whether system has occurred failure；And another kind of is based on number
According to the method for driving, this method is strongly depend on the size and performance of data set, needs a large amount of historical data, and according to statistics
The confidence limit to calculate them is analyzed, such method is represented as PCA method.Currently used modeling method has autoregression sliding
The method of average, support vector machines (SVM), Kalman filtering (KF), gauss hybrid models (GMM), recurrence least square (RLS), base
In the model etc. of core filtering algorithm, such methods assume that data meet Gaussian Profile and only consider current time control information mostly
Bring influences, and cannot efficiently differentiate the exception of the quick variation and sensor signal of blood glucose level data, furthermore above-mentioned to be previously mentioned
These methods mostly use constant threshold value greatly, however blood glucose level data is dynamic change, and constant threshold value will lead to detection system
It unites insensitive to some slowly varying fault-signals and some glitch signals.
Summary of the invention
(1) technical problems to be solved
For existing technical problem, the present invention provides a kind of continuous blood sugar monitoring sensing based on multi-model fusion
The online fault detection method of device solves the problems such as accuracy rate of testing result is low in the prior art.
(2) technical solution
In order to achieve the above object, the main technical schemes that the present invention uses include:
A kind of continuous blood sugar monitoring online fault detection method of sensor based on multi-model fusion, includes the following steps:
S1, online CGM monitoring signals data are obtained；
In S2, the online CGM monitoring signals data that will acquire input multi-model blending algorithm model, obtains on-line prediction and miss
Difference；
S3, the on-line prediction error that will acquire and historical forecast error combine the entropy for calculating and obtaining the online moment；
S4, the entropy J that the online moment obtained will be calculatedi1、Ji2Respectively with the threshold value T at current timekl1、Tkl2Compare；
If the entropy J at current timei1、Ji2It is not greater than the threshold value T at current time entirelykl1、Tkl2, then judge that current blood glucose is supervised
It is normal to survey working sensor；
If the entropy J at current timei1、Ji2It is all larger than the threshold value T at current timekl1、Tkl2, then judge that current blood glucose monitors
Working sensor is abnormal.
Preferably, when judging that current blood glucose monitoring working sensor is normal, the method also includes:
S40A1, secondary detection；
S40A2, return step S1.
Preferably, when judging that current blood glucose monitors working sensor exception, the method also includes:
S40B1, return step S1；
S40B2, the predicted value of current multi-model blending algorithm model is replaced into the monitoring signals data in step 1.
Preferably, the step S40A1 further include:
A1, two entropy J for calculating current timei1And Ji2；
A2, by two entropy Ji1、Ji2Respectively with threshold value Tkl1、Tkl2It is compared；
If entropy Ji1And Ji2It is more than threshold value Tkl1、Tkl2；
Then judge that current blood glucose monitoring working sensor is abnormal, replace measured value with the predicted value of model, and to it is current when
The model predictive error at quarter is reconstructed；
Otherwise judge that current blood glucose monitoring working sensor is normal, be updated using parameter of the measured value to model.
Preferably, judge that after current blood glucose monitors working sensor exception further include following steps in the step A2:
B1, the entropy J for calculating current time1And J2；
B2, the entropy J for judging current time1And J2Value whether rise, if rise, execute B3, otherwise execute B4；
B3, judge whether last moment measured value is fault value, if it is not, then given threshold is 3 δ confidences of entropy in window
Section, and the entropy at current time and the entropy of last moment are stored, otherwise update storage the entropy at current time；
B4, judge whether last moment measured value is fault value, if so, when given threshold is that nearest moment entropy declines
The 95% of maximum changing value, otherwise threshold value is reasonable.
Preferably, the entropy J at current time is calculated in the step B11And J2Further include following steps:
The model predictive error of C1, the multi-model blending algorithm model at acquisition current time；
C2, the model predictive error for obtaining historical juncture and nearest moment；
C3, the mean value and variance for calculating separately historical juncture and nearest moment model predictive error, as p (x) and p1(x)
The mean value and variance of distribution；
C4, calculate separately comprising after current time model predictive error historical juncture and nearest moment model predictive error
Mean value and variance, as q (x) and q1(x) mean value and variance being distributed；
C5, using the calculation formula of KL divergence, calculate the entropy J at current time1And J2；
Preferably, the calculation formula of the KL divergence in the step C5 are as follows:
Wherein, p (x) and q (x) is two single argument normal distributions, and meets p~N (μ0, σ0) and q~N (μ1, σ1)。
Preferably, the multi-model blending algorithm model includes the following steps:
D1, the blood glucose level data g (t) that continuous blood sugar monitoring sensor measurement arrives is obtained；
D2, the data of acquisition are reconstructed using the sliding window that a length is L, obtain the input square of following form
Battle array and output matrix:
Y (N*1) two [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 are the sample numbers for predicting, K is former
The sample number of beginning time series, PH indicate the step number of advanced prediction, and i indicates current time；
D3, the data after reconstruct are modeled respectively using SVM and RLS algorithm, generates the predicted value y of modelSVMWith
yRLS；
D4, the mean value for calculating each model history prediction error；
D5, judge whether the model predictive error of each model last moment is greater than 3, if so, executing step D6, otherwise hold
Row step D7；
D6, compare two model predictive error meansvmAnd meanrlsSize, if meansvm< meanrls, then model
Final predicted value be Y=ySVM, otherwise Y=yRLS’, and terminate；
D7, the prediction error according to each model, calculate the weight of each model；
D8, the expression formula such as following formula for calculating the fused model predication value of multi-model:
Y=ySVM×wsvm+yRLS×wrls
Wherein, wsvm+vrls=1.
Preferably, the calculation formula that the mean value of the historical forecast error of each model is calculated in the step D4 is as follows:
Wherein, errorsvmAnd errorrlsThe respectively prediction error of SVM and RLS model, n are the number of model predictive error
It measures and meets i > n；
Preferably, the formula that each Model Weight is calculated in the step D7 is as follows:
Wherein, wsvmFor weight shared by SVM model；wrlsFor weight shared by RLS model.
(3) beneficial effect
The beneficial effects of the present invention are: a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors sensor
Online fault detection method, the historical data according to acquisition generate the model prediction at current time by multi-model fusion method
Value, and analyzed using prediction error of the history error of model prediction to current time, while considering the dynamic of blood glucose level data
State property designs a kind of dynamic threshold more new strategy, can effectively distinguish the exception of the quick variation and sensor signal of blood glucose level data.
In addition, the present invention can be handled fault-signal, it can effectively avoid insulin pump and infused according to the blood glucose level data information of mistake
Enter the insulin of number of errors, and then reduction is influenced caused by patient vitals' safety.
The predictive ability of model can be improved in multi-model fusion forecasting method, reduces model predictive error to resulting
It influences, it is contemplated that the dynamic of blood glucose level data, the present invention propose a kind of multimode based on recurrence least square and support vector machines
Type fusion forecasting method is used to improve the precision of prediction of model.Wherein, recursive least square method fast convergence rate, can be quick
Ground follows the fluctuation of blood glucose level data, and support vector machine method is then that nonlinear data is mapped to height by inner product kernel function
Dimension space is converted into linear data and is handled, it can preferable fit non-linear data, but fluctuating biggish extreme value
Point nearby, predicts that error is larger, therefore the multi-model prediction technique that is blended based on above-mentioned analysis using two methods is improved
The precision of prediction of model.
Detailed description of the invention
Fig. 1 is that a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors the online fault detection side of sensor
The flow diagram of method；
Fig. 2 is that a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors the online fault detection side of sensor
The flow diagram of multi-model blending algorithm model in method；
Fig. 3 is that a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors the online fault detection side of sensor
The flow diagram that dynamic threshold updates in method；
Fig. 4 is that a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors the online fault detection side of sensor
The algorithm flow schematic diagram of method；
Fig. 5 is that a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors the online fault detection side of sensor
The algorithm flow schematic diagram of multi-model blending algorithm model in method；
Fig. 6 is that a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors the online fault detection side of sensor
The comparing result schematic diagram that different models predict blood glucose level data in method；
Fig. 7 is that a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors the online fault detection side of sensor
Online fault detection schematic diagram in method based on KL divergence；
Fig. 8 is that a kind of continuous blood sugar based on multi-model fusion provided by the invention monitors the online fault detection side of sensor
The schematic diagram of calculation flow of KL divergence in method.
Specific embodiment
In order to preferably explain the present invention, in order to understand, with reference to the accompanying drawing, by specific embodiment, to this hair
It is bright to be described in detail.
As shown in Figure 1 and Figure 4: present embodiment discloses a kind of, and the continuous blood sugar monitoring sensor based on multi-model fusion exists
Line fault detection method, includes the following steps:
S1, online CGM monitoring signals data are obtained；
In S2, the online CGM monitoring signals data that will acquire input multi-model blending algorithm model, obtains on-line prediction and miss
Difference；
S3, the on-line prediction error that will acquire and historical forecast error combine the entropy for calculating and obtaining the online moment；
S4, the entropy J that the online moment obtained will be calculatedi1、Ji2Respectively with the threshold value T at current timekl1、Tkl2Compare；
If the entropy J at current timei1、Ji2It is not greater than the threshold value T at current time entirelykl1、Tkl2, then judge that current blood glucose is supervised
It is normal to survey working sensor；
If the entropy J at current timei1、Ji2It is all larger than the threshold value T at current timekl1、Tkl2, then judge that current blood glucose monitors
Working sensor is abnormal.
In the present embodiment when judging that current blood glucose monitoring working sensor is normal, the method also includes:
S40A1, secondary detection；
S40A2, return step S1.
In the present embodiment when judging that current blood glucose monitors working sensor exception, the method also includes:
S40B1, return step S1；
S40B2, the predicted value of current multi-model blending algorithm model is replaced into the monitoring signals data in step 1.
Step S40A1 described in the present embodiment further include:
A1, two entropy J for calculating current timei1And Ji2；
A2, by two entropy Ji1、Ji2Respectively with threshold value Tkl1、Tkl2It is compared；
If entropy Ji1And Ji2It is more than threshold value Tkl1、Tkl2；
Then judge that current blood glucose monitoring working sensor is abnormal, replace measured value with the predicted value of model, and to it is current when
The model predictive error at quarter is reconstructed；
Otherwise judge that current blood glucose monitoring working sensor is normal, be updated using parameter of the measured value to model.
Here secondary detection can check whether the work of sensor is abnormal, greatly improves inspection more fully hereinafter
Survey the accuracy of result.
It is as shown in Figure 3: also to be wrapped after judging current blood glucose monitoring working sensor exception in step A2 described in the present embodiment
Include following steps:
B1, the entropy J for calculating current time1And J2；
B2, the entropy J for judging current time1And J2Value whether rise, if rise, execute B3, otherwise execute B4；
B3, judge whether last moment measured value is fault value, if it is not, then given threshold is 3 δ confidences of entropy in window
Section, and the entropy at current time and the entropy of last moment are stored, otherwise update storage the entropy at current time；
B4, judge whether last moment measured value is fault value, if so, when given threshold is that nearest moment entropy declines
The 95% of maximum changing value, otherwise threshold value is reasonable.
Dynamic threshold provided herein more new strategy can effectively distinguish the quick variation and sensor signal of blood glucose level data
It is abnormal.
It is as shown in Figure 8: the entropy J at current time is calculated in step B1 described in the present embodiment1And J2It further include walking as follows
It is rapid:
The model predictive error of C1, the multi-model blending algorithm model at acquisition current time；
C2, the model predictive error for obtaining historical juncture and nearest moment；
C3, the mean value and variance for calculating separately historical juncture and nearest moment model predictive error, as p (x) and p1(x)
The mean value and variance of distribution；
C4, calculate separately comprising after current time model predictive error historical juncture and nearest moment model predictive error
Mean value and variance, as q (x) and q1(x) mean value and variance being distributed；
C5, using the calculation formula of KL divergence, calculate the entropy J at current time1And J2；
The calculation formula of KL divergence in step C5 described in the present embodiment are as follows:
Wherein, p (x) and q (x) is two single argument normal distributions, and meets p~N (μ0, σ0) and q~N (μ1, σ1)。
As shown in Figure 2 and Figure 5: the model of multi-model blending algorithm described in the present embodiment includes the following steps:
D1, the blood glucose level data g (t) that continuous blood sugar monitoring sensor measurement arrives is obtained；
D2, the data of acquisition are reconstructed using the sliding window that a length is L, obtain the input square of following form
Battle array and output matrix:
Y (N*1) two [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 are the sample numbers for predicting, K is former
The sample number of beginning time series, PH indicate the step number of advanced prediction, and i indicates current time；
D3, the data after reconstruct are modeled respectively using SVM and RLS algorithm, generates the predicted value y of modelSVMWith
yRLS；
D4, the mean value for calculating each model history prediction error；
D5, judge whether the model predictive error of each model last moment is greater than 3, if so, executing step D6, otherwise hold
Row step D7；
D6, compare two model predictive error meansvmAnd meanrlsSize, if meansvm< meanrls, then model
Final predicted value be Y=ySVM, otherwise Y=yRLS’, and terminate；
D7, the prediction error according to each model, calculate the weight of each model；
D8, the expression formula such as following formula for calculating the fused model predication value of multi-model:
Y=ySVM×wsvm+yRLS×wrls
Wherein, wsvm+wrls=1.
Multi-model blending algorithm model is generated according to the historical data of acquisition by multi-model fusion method in the present embodiment
The model predication value at current time, and analyzed using prediction error of the history error of model prediction to current time, it examines
The result of survey is more accurate.
It is noted that the following institute of the calculation formula for calculating the mean value of the historical forecast error of each model in the step D4
Show:
Wherein, errorsvmAnd errorrlsThe respectively prediction error of SVM and RLS model, n are the number of model predictive error
It measures and meets i > n；
It is noted that the formula for calculating each Model Weight in the step D7 is as follows:
Wherein, wsvmFor weight shared by SVM model；wrlsFor weight shared by RLS model.
1. the performance verification of multi-model fusion forecasting method
The method proposed in the present embodiment is in U.S. Fei Jiniya university/Padova, Italy university type-1 diabetes mellitus generation
It thanks and is verified on simulator.Experimental data was obtained for interval by sampling with five minutes, included adult, blueness in data set
Juvenile and 6 days blood glucose level datas of children's three classes patient, by taking the method for one-step prediction to assess the performance of each model,
As a result as shown in Figure 6.The model prediction performance comparison using single model and multi-model fusion method is given in table 1 as a result,
By analyzing each model to the prediction result of three classes glucose data, it can be deduced that more compared to single model prediction method
Degree of fitting between Model Fusion prediction technique and blood glucose level data is preferable, to illustrate the reasonability of algorithm proposed by the present invention.
The comparison of the estimated performance of the single model of table 1 and multi-model hybrid forecasting method
2. the performance verification of online fault detection algorithm
The common fault type of continuous blood sugar monitoring sensor has six kinds, is spike, drift, step, pressure sensitive respectively
Sensor decaying, dropout and stagnation.Wherein dropout and stagnation are easily detected out, small with 4 in simulation process
When be period random addition remaining four kinds of fault-signal into normal blood glucose level data, the amplitude of fault-signal is set as
10%, Fig. 7 are failure detection result of the online fault detection algorithm on children's blood glucose level data.It can be seen from the figure that this hair
The detection method of bright middle proposition can timely and effectively detect the glitch signal of four seed type failures, and can be to fault-signal
It is reconstructed, to reduce harm caused by when continuous blood sugar monitoring sensor breaks down, continuous blood sugar prison is effectively promoted
Survey the performance of sensor.
The technical principle of the invention is described above in combination with a specific embodiment, these descriptions are intended merely to explain of the invention
Principle shall not be construed in any way as a limitation of the scope of protection of the invention.Based on explaining herein, those skilled in the art
It can associate with other specific embodiments of the invention without creative labor, these modes fall within this hair
Within bright protection scope.
Claims (10)
1. a kind of continuous blood sugar based on multi-model fusion monitors the online fault detection method of sensor, which is characterized in that including
Following steps:
S1, online CGM monitoring signals data are obtained；
In S2, the online CGM monitoring signals data that will acquire input multi-model blending algorithm model, on-line prediction error is obtained；
S3, the on-line prediction error that will acquire and historical forecast error combine the entropy for calculating and obtaining the online moment；
S4, the entropy J that the online moment obtained will be calculatedi1、Ji2Respectively with the threshold value T at current timekl1、Tkl2Compare；
If the entropy J at current timei1、Ji2It is not greater than the threshold value T at current time entirelykl1、Tkl2, then judge current blood glucose monitoring sensing
Device is working properly；
If the entropy J at current timei1、Ji2It is all larger than the threshold value T at current timekl1、Tkl2, then judge that current blood glucose monitors sensor
Operation irregularity.
2. detection method according to claim 1, which is characterized in that judging that current blood glucose monitoring working sensor is normal
When, the method also includes:
S40A1, secondary detection；
S40A2, return step S1.
3. detection method according to claim 2, which is characterized in that judging that current blood glucose monitoring working sensor is abnormal
When, the method also includes:
S40B1, return step S1；
S40B2, the predicted value of current multi-model blending algorithm model is replaced into the monitoring signals data in step 1.
4. detection method according to claim 2, which is characterized in that the step S40A1 further include:
A1, two entropy J for calculating current timei1And Ji2；
A2, by two entropy Ji1、Ji2Respectively with threshold value Tkl1、Tkl2It is compared；
If entropy Ji1And Ji2It is more than threshold value Tkl1、Tkl2；
Then judge that current blood glucose monitoring working sensor is abnormal, replaces measured value with the predicted value of model, and to current time
Model predictive error is reconstructed；
Otherwise judge that current blood glucose monitoring working sensor is normal, be updated using parameter of the measured value to model.
5. detection method according to claim 4, which is characterized in that judge current blood glucose monitoring sensing in the step A2
Further include following steps after device operation irregularity:
B1, the entropy J for calculating current time1And J2；
B2, the entropy J for judging current time1And J2Value whether rise, if rise, execute B3, otherwise execute B4；
B3, judge whether last moment measured value is fault value, if it is not, then given threshold is 3 δ confidence areas of entropy in window
Between, and the entropy at current time and the entropy of last moment are stored, otherwise update storage the entropy at current time；
B4, judge whether last moment measured value is fault value, if so, given threshold is that nearest moment entropy is maximum when declining
The 95% of changing value, otherwise threshold value is reasonable.
6. detection method according to claim 5, which is characterized in that calculate the entropy J at current time in the step B11
And J2Further include following steps:
The model predictive error of C1, the multi-model blending algorithm model at acquisition current time；
C2, the model predictive error for obtaining historical juncture and nearest moment；
C3, the mean value and variance for calculating separately historical juncture and nearest moment model predictive error, as p (x) and p1(x) it is distributed
Mean value and variance；
C4, calculate separately comprising after current time model predictive error historical juncture and nearest moment model predictive error it is equal
Value and variance, as q (x) and q1(x) mean value and variance being distributed；
C5, using the calculation formula of KL divergence, calculate the entropy J at current time1And J2。
7. detection method according to claim 6, which is characterized in that the calculation formula of the KL divergence in the step C5
Are as follows:
Wherein, p (x) and q (x) is two single argument normal distributions, and meets p~N (μ0, σ0) and q~N (μ1, σ1)。
8. detection method according to claim 1-7, which is characterized in that the multi-model blending algorithm model packet
Include following steps:
D1, the blood glucose level data g (t) that continuous blood sugar monitoring sensor measurement arrives is obtained；
D2, using a length be L sliding window the data of acquisition are reconstructed, obtain following form input matrix and
Output matrix:
Y (N*1)=[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 sample number for predicting, when K is original
Between sequence sample number, PH indicate advanced prediction step number, i indicate current time；
D3, the data after reconstruct are modeled respectively using SVM and RLS algorithm, generates the predicted value y of modelSVMAnd yRLS；
D4, the mean value for calculating each model history prediction error；
D5, judge whether the model predictive error of each model last moment is greater than 3, if so, executing step D6, otherwise execute step
Rapid D7；
D6, compare two model predictive error meansvm: and meanrlsSize, if meansvm< meanrls, then model
Final predicted value is Y=ySVM, otherwise Y=yRLS, and terminate；
D7, the prediction error according to each model, calculate the weight of each model；
D8, the expression formula such as following formula for calculating the fused model predication value of multi-model:
Y=ySVM×wsvm+yRLS×wrls
Wherein, wsvm+wrls=1.
9. detection method according to claim 8, which is characterized in that calculate the historical forecast of each model in the step D4
The calculation formula of the mean value of error is as follows:
Wherein, errorsvm: and errorrlsThe respectively prediction error of SVM and RLS model, n are the quantity of model predictive error
And meet i > n.
10. detection method according to claim 8, which is characterized in that calculate the public affairs of each Model Weight in the step D7
Formula is as follows:
Wherein, wsvmFor weight shared by SVM model；wrlsFor weight shared by RLS model.
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 true CN109682976A (en) | 2019-04-26 |
| CN109682976B 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) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110379503A (en) * | 2019-07-30 | 2019-10-25 | 东北大学 | A kind of online fault detection and diagnosis system based on continuous blood sugar monitoring system |
| CN110448306A (en) * | 2019-07-30 | 2019-11-15 | 东北大学 | A kind of online fault detection and diagnosis method based on continuous blood sugar monitoring system |
| CN110490266A (en) * | 2019-08-23 | 2019-11-22 | 北京邮电大学 | A kind of sensing data uploads, Transducer-fault Detecting Method and device |
| 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 |
| CN112750333A (en) * | 2019-10-30 | 2021-05-04 | 株式会社日立制作所 | Abnormality determination device, vehicle assistance system, and server |
| EP3996590A4 (en) * | 2019-07-10 | 2023-08-02 | University Of Virginia Patent Foundation | System and method for online domain adaptation of models for hypoglycemia prediction in type 1 diabetes |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060017923A1 (en) * | 1999-01-22 | 2006-01-26 | Ruchti Timothy L | Analyte filter method and apparatus |
| 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 | A kind of abnormal fasting blood sugar method for early warning based on integrated study Fusion Model |
-
2019
- 2019-02-28 CN CN201910153512.4A patent/CN109682976B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060017923A1 (en) * | 1999-01-22 | 2006-01-26 | Ruchti Timothy L | Analyte filter method and apparatus |
| 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 | A kind of abnormal fasting blood sugar method for early warning based on integrated study Fusion Model |
Non-Patent Citations (4)
| Title |
|---|
| JIANYUAN FENG，ET AL.: "Hybrid Online Sensor Error Detection and Functional Redundancy for Artificial Pancreas Control Systems", 《11TH IFAC SYMPOSIUM ON DYNAMICS AND CONTROL OF PROCESS SYSTEMS,INCLUDING BIOSYSTEMS》 * |
| KAMURAN TURKSOY，ET AL.: "Monitoring and Fault Detection of Continuous Glucose Sensor Measurements", 《AMERICAN CONTROL CONFERENCE》 * |
| XIA YU，ET AL.: "Fault Detection in Continuous Glucose Monitoring Sensors for Artifical Pancreas Systems", 《IFAC PAPERSONLINE》 * |
| XIA YU，ET AL.: "Model-Fusion-Based Online Glucose Concentration Predictions in People with Type 1 Diabetes", 《CONTROL ENG PRACT》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3996590A4 (en) * | 2019-07-10 | 2023-08-02 | University Of Virginia Patent Foundation | System and method for online domain adaptation of models for hypoglycemia prediction in type 1 diabetes |
| CN110379503A (en) * | 2019-07-30 | 2019-10-25 | 东北大学 | A kind of online fault detection and diagnosis system based on continuous blood sugar monitoring system |
| CN110448306A (en) * | 2019-07-30 | 2019-11-15 | 东北大学 | A kind of online fault detection and diagnosis method based on continuous blood sugar monitoring system |
| 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 |
| CN110490266A (en) * | 2019-08-23 | 2019-11-22 | 北京邮电大学 | A kind of sensing data uploads, Transducer-fault Detecting Method and device |
| CN112750333A (en) * | 2019-10-30 | 2021-05-04 | 株式会社日立制作所 | Abnormality determination device, vehicle assistance system, and server |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109682976B (en) | 2020-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109682976A (en) | Continuous blood sugar based on multi-model fusion monitors the online fault detection method of sensor | |
| CN107463633B (en) | A kind of real time data rejecting outliers method based on EEMD- neural network | |
| CN108446523A (en) | A kind of assessment of complete electronic set storage life and prediction technique | |
| CN111033413B (en) | Monitoring device and method for monitoring a system | |
| CN111314173B (en) | Monitoring information abnormity positioning method and device, computer equipment and storage medium | |
| CN105403705B (en) | Continuous blood glucose monitoring equipment comprising blood glucose classification function fault detection module | |
| Staal et al. | Kalman smoothing for objective and automatic preprocessing of glucose data | |
| CN104986347B (en) | Real-time detection method for civil aircraft airline pilot operation errors | |
| CN109254219B (en) | A kind of distribution transforming transfer learning method for diagnosing faults considering multiple factors Situation Evolution | |
| CN110448306B (en) | Online fault detection and diagnosis method based on continuous blood glucose monitoring system | |
| CN110379503B (en) | Online fault detection and diagnosis system based on continuous blood glucose monitoring system | |
| CN106528313A (en) | Host variable abnormity detection method and system | |
| CN112052496B (en) | Operation method of high arch dam valley amplitude deformation influence factor analysis system based on VAR model | |
| JP5949032B2 (en) | Pre-processing method and abnormality diagnosis device | |
| CN110841152A (en) | Heatable pressure device for blood transfusion of hematology department | |
| US11244235B2 (en) | Data analysis device and analysis method | |
| CN105117615A (en) | Anomaly detection method based on cumulative sum control chart and applied to satellite power supply system | |
| CN106815468B (en) | A kind of On-line Estimation method from confirmation air data system uncertainty of measurement | |
| CN115310705A (en) | Method and device for determining gas emission quantity and computer readable storage medium | |
| CN112205974A (en) | Intelligent blood pressure management analysis system based on LSTM model | |
| CN117766145A (en) | Continuous blood glucose prediction model construction method, blood glucose prediction method and device | |
| CN105388276B (en) | A kind of Blood Kit analysis system and analysis method | |
| CN117079760B (en) | Warfarin dosage adjustment method, warfarin dosage adjustment system and clinical auxiliary decision-making system | |
| Zhao et al. | Automatic and online fault detection of sensor problems using continuous glucose monitoring data for type 1 diabetes | |
| CN110364240A (en) | Health data monitoring method, device, computer equipment and storage medium |
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 |