CN103310113B - A kind ofly to be separated and the general blood glucose prediction method of data modeling based on frequency band - Google Patents

Abstract

The invention discloses and to be a kind ofly separated and the general blood glucose prediction method of data modeling based on frequency band, the method is analyzed for the subcutaneous blood glucose measurement signal of human body, extract its potential time-series dynamics characteristic, and subcutaneous blood glucose signal is divided into height two frequency ranges by the threshold value defining frequency band separation; Analyze its sequential autocorrelation for low frequency blood glucose signal, set up autoregression blood glucose prediction model; General blood glucose prediction method proposed by the invention for new object without the need to etc. re-start modeling after abundant blood glucose measurement signal to be obtained, but the forecast model that directly can call other individuality carries out real time blood sugar prediction, enormously simplify modeling work amount and complexity, greatly will reduce modeling cost; And universal model adopts the method based on frequency band separation and latent variable modeling, improves precision of prediction.The present invention is easy to implement, and the research for blood glucose prediction modeling method specifies new direction.

Description

A kind ofly to be separated and the general blood glucose prediction method of data modeling based on frequency band

Technical field

The invention belongs to blood glucose level data analysis and prediction research field, particularly relate to and to be a kind ofly separated and the general blood glucose prediction method of data modeling based on frequency band.

Background technology

A distinguishing feature of blood sugar for human body level is exactly time variation, is embodied in clock signal measured value and there is significant auto-correlation relation.Carry out analyzing for blood glucose signal, modeling, its sequential correlation properties can be extracted, obtain following change of blood sugar situation according to historical glucose dynamic.Foreign scholar Bremer and Gough in 1999 proposes blood sugar time series data first and has a kind of potential dependency structure, can be described by the simple linear dynamic model of one.At present, the foundation of blood glucose prediction model adopts the method based on data-driven more.Already present forecast model can be divided into linearly (to be its Typical Representative based on the autoregressive model of least square) and non-linear (be its Typical Representative with neural net method) two classes.Linear model, due to its simple model structure and algorithm, is applied widely.The modeling method of comparative maturity comprises autoregression, exciter response (impulse-response, IR) etc.Autoregression method is the Data Modeling Method of comparative maturity, and it only utilizes the information of blood sugar itself, obtains following blood glucose prediction value by the linear combination of historical glucose data.But traditional autoregressive model major defect has 2 points: (a) directly utilizes least square the most basic this discrimination method matching blood glucose prediction relation for measurement data, the defect that the method itself has cannot be avoided, satisfied precision of prediction can not be obtained; B () is not analyzed in advance for the blood sugar dynamic of Different Individual, directly set up different forecast models (i.e. individuation model) for on-line prediction, this is by substantial manpower and materials.

Summary of the invention

The object of the invention is to the deficiency for existing blood glucose prediction method, provide a kind of and be separated and the general blood glucose prediction method of data modeling based on frequency band.

The object of the invention is to be achieved through the following technical solutions: to be a kind ofly separated based on frequency band and the general blood glucose prediction method of data modeling, the method comprises the following steps:

Step 1: modeling blood glucose signal pre-service: the subcutaneous blood glucose signal of individuality obtained with certain sampling period Δ t is combined into one dimension time series data x ^t(1 × Z), wherein, x is the measured value of blood glucose signal, and Z is number of samples, removes spike noise wherein.Timing dependence and the dynamic-change information of blood glucose signal is contained in this one dimension time series data.

Step 2: blood glucose signal frequency band is separated: the Changing Pattern of dynamic in high and low different frequency range analyzing blood sugar, distinguishes critical frequencies section and secondary frequency band, determines the optimal threshold that frequency band divides.Butterworth LPF is adopted to carry out frequency band separation to blood glucose signal according to the separation threshold value of definition.

Step 3: obtain predictive variable matrix and response matrix: slip over x with the one dimension moving window that a length is K sampled point ^lT(1 × Z), each mobile sampled point, mobile Z-K+1 time altogether.Using the data in each moving window as a new row vector, then can be combined as a two-dimensional data matrix X ^l(N × K), wherein, N=Z-K+1, K=PL+H (PH/5).PL represents the length of predictive variable, and H represents prediction step, and PH represents forecast interval, H with PH linear module is different, but all represent predict be how many walk after blood sugar.Because prediction step is often walk 5 minutes, therefore there is PH=5 × H.

Step 4: the blood glucose prediction modeling be separated based on frequency band: carry out blood glucose prediction modeling by the method based on latent variable.

Step 5: predict according to the future blood glucose value of blood glucose prediction model to any individual that step 4 is set up.Come by following steps:

(5.1), during on-line prediction, any individual new data is being collected afterwards (subscript n ew represents new samples, J=PL), right the frequency band separating treatment of carrying out as described in step 2 obtains

(5.2) call general low frequency model and carry out on-line prediction:

(5.2.1) call the general low frequency model set up based on least square and carry out on-line prediction: for each new test data the predicted value after PH sampled point is generated by following steps

${\hat{y}}_{new}^L=x_{new}^{LT}{\mathrm{\θ}}^L;$

Wherein, θ ^lfor above according to the regression coefficient vector that training data is obtained.

(5.2.2) call the general low frequency model set up based on latent variable and carry out on-line prediction:

For each new test data the predicted value after obtaining PH sampled point is calculated by following steps

${\hat{y}}_{new}^L=x_{new}^{LT}{\mathrm{\σ}}^L;$

Wherein, σ ^lit is the regression coefficient vector that offset minimum binary-Canonical Correlation Analysis is obtained.

(5.3) precision of prediction is weighed: the prediction of quality result of acquisition and actual measured value are contrasted.After a series of new prediction signal of acquisition, can according to the blood glucose prediction precision of the RMSE defined in following formula (root-mean-square error) performance Index Calculation for new samples:

$RMSE=\sqrt{\frac{1}{N}\underset{i\∈N}{\Σ}{\left(y\right(i)-{\hat{y}}^{*}(i\left)\right)}^2};$

Wherein, y (i) represents original blood sugar measured, represent the blood glucose prediction result adopting different model to obtain, adopt the low frequency blood glucose prediction value that the first same individual model or Different Individual model are tried to achieve or adopt the full range predicted value that the second same individual model is tried to achieve n is sample number.

Compared with prior art, the invention has the beneficial effects as follows: general blood glucose prediction method proposed by the invention for new object without the need to etc. re-start modeling after abundant blood glucose measurement signal to be obtained, but the forecast model that directly can call other individuality carries out real time blood sugar prediction, enormously simplify modeling work amount and complexity, greatly will reduce modeling cost; And universal model adopts the method based on frequency band separation and latent variable modeling, improves precision of prediction.The present invention is easy to implement, and the research for blood glucose prediction modeling method specifies new direction.

Accompanying drawing explanation

Fig. 1 is general blood glucose prediction model modeling process flow diagram of the present invention;

Fig. 2 is the comparison diagram of height frequency range blood glucose signal after the subcutaneous blood glucose signal of full range provided by CGM (continuous blood sugar monitor) is separated with frequency band;

Fig. 3 is the effect diagram (ordinate be the mean value of RMSE) of PL to general low frequency model (GL) blood glucose prediction performance;

Fig. 4 is the effect diagram (ordinate be the MAD value of RMSE) of PL to general low frequency model (GL) blood glucose prediction performance;

Fig. 5 is the effect diagram (ordinate be the mean value of RMSE) of PH to general low frequency model (GL) blood glucose prediction performance;

Fig. 6 is the effect diagram (ordinate be the MAD value of RMSE) of PH to general low frequency model (GL) blood glucose prediction performance;

Fig. 7 is the blood glucose measurement curve of general low frequency (GL) model and the comparison diagram (analytic target is first group of the first blood glucose signal) of blood glucose prediction curve.

Fig. 8 is the blood glucose measurement curve of standard independent individual (SD) model and the comparison diagram (analytic target is first group of the first blood glucose signal) of blood glucose prediction curve.

Fig. 9 is the blood glucose measurement curve of general low frequency (GL) model and the comparison diagram (analytic target is second group of the first blood glucose signal) of blood glucose prediction curve.

Figure 10 is the blood glucose measurement curve of standard independent individual (SD) model and the comparison diagram (analytic target is second group of the first blood glucose signal) of blood glucose prediction curve.

Embodiment

As shown in Figure 1, the present invention is based on the general blood glucose prediction method of frequency band separation and data modeling, the method comprises the following steps:

Step 1: modeling blood glucose signal pre-service

For the subcutaneous blood glucose signal of individuality (here Δ t=5min) obtained with certain sampling period Δ t, one dimension time series data x can be combined into ^t(1 × Z), wherein, x is the measured value of blood glucose signal, and Z is number of samples, removes spike noise wherein.Timing dependence and the dynamic-change information of blood glucose signal is contained in this one dimension time series data.In this example, we have the blood sugar clock signal coming from two group objects, and the 1st group comprises 12 people, and the 2nd group comprises 14 people, two groups of totally 26 people.The blood sugar clock signal of each object comprises the data of two or three days.The data of first day are used for model validation, and remainder data is used for the checking of model performance.

Step 2: blood glucose signal frequency band is separated

The Changing Pattern of dynamic in high and low different frequency range of this step Main Analysis blood sugar, distinguishes critical frequencies section and secondary frequency band, determines the optimal threshold that frequency band divides.Butterworth LPF is adopted to carry out frequency band separation to blood glucose signal according to the separation threshold value of definition.Fundamental purpose is separated by frequency band, removes the noise effect of high band, retains effective change of blood sugar information of low-frequency range.

(2.1) the single order low pass Butterworth filter that form of choosing is following:

$\tilde{x}\left(k\right)={\mathrm{\β}}_{1}x\left(k\right)+{\mathrm{\β}}_{2}x(k-1)-\mathrm{\α}\tilde{x}\left(k\right);---\left(1\right)$

Arranging its threshold values cycle is P (min).Wherein, x is the measured value of blood glucose signal, its filter value, α, β ₁and β ₂be filtering parameter, k is sampled point.Therefore filtering output is the linear combination of history filter value, history and current measurement value.According to the understanding of reality to blood sugar dynamic, P value is generally 40min ~ 80min.

(2.2) to the data x gathered in step 1 ^t(1 × Z) carries out filtering by single order Butterworth LPF, isolates low frequency blood glucose level data x ^lT(1 × Z) and high frequency blood glucose level data x ^hT(1 × Z).High frequency blood glucose level data is considered to noise, can remove, and step is:

x ^LT(1×Z)＝x ^T(1×Z)-x ^HT(1×Z)；(2)

Wherein, subscript L, H represents low frequency and high-frequency data respectively.

Step 3: obtain predictive variable matrix and response matrix:

X is slipped over the one dimension moving window that a length is K sampled point ^lT(1 × Z), each mobile sampled point, mobile Z-K+1 time altogether.Using the data in each moving window as a new row vector, then can be combined as a two-dimensional data matrix X ^l(N × K), wherein, N=Z-K+1, K=PL+H (PH/5).PL (PredictorLength) represents the length of predictive variable, H represents prediction step, PH (PredictionHorizon) represents forecast interval, H with PH linear module is different, but all represents that what predict is blood sugar after how many steps (generally often step for 5min).Because prediction step in the present invention is often walk 5 minutes, therefore there is PH=5 × H.

Step 4: the blood glucose prediction modeling be separated based on frequency band:

Forecast model is set up two kinds of methods, and a kind of is the existing method based on least square, and another kind is the modeling method based on latent variable.Concrete steps are as follows:

(4.1) predictive variable and output variable is chosen: according to the value of PL, choose x ^l(N × K) front PL arranges as prediction matrix X ^l(N × J), J=PL; Then X ^l(N × K) PL+PH arranges as output variable y ^l(N × 1).

If (4.2.) carry out blood glucose prediction modeling by the method based on least square, then adopt latent variable method establishment low frequency Self-regression Forecast Model according to low frequency blood glucose signal:

$y^L=X^L{\mathrm{\θ}}^L+f={\hat{y}}^L+f;---\left(3\right)$

Wherein, for the low frequency blood glucose prediction value obtained, θ ^l(J × 1) is regression coefficient, and f (N × 1) is model error.

If carry out blood glucose prediction modeling by the method based on latent variable, adopt latent variable method establishment Self-regression Forecast Model according to low frequency blood glucose signal:

Autoregression low frequency forecast model based on latent variable adopts the feature extracting method of offset minimum binary-canonical correlation analysis to carry out modeling.First extract preliminary latent variable group with offset minimum binary, the latent variable latent variable extracted due to deflected secondary air can not ensure the closely related relation of itself and response variable, with canonical correlation analysis, aftertreatment is carried out to it again, thus extract wherein with relevant variable closely-related part carry out regression modeling and prediction.Step is as follows:

(4.2.1) data prediction

The predictive variable matrix of all individualities and response matrix are grouped together, for the variable x of any point in the predictive variable matrix after combination and response matrix _i,j, average, global criteria process divided by standard deviation are subtracted to this variable, wherein, subscript i representative batch, j represents variable, and the computing formula of standardization is as follows:

$x_{i,j}=\frac{x_{i,j}-{\stackrel{\‾}{x}}_j}{s_j};---\left(4\right)$

Wherein: the average of the arbitrary row of matrix after combination, s _jit is the standard deviation of respective column.Its computing formula is as follows:

$\begin{array}{c}{\stackrel{\‾}{x}}_j=\frac{1}{M}\underset{i=1}{\overset{M}{\Σ}}{x}_{i,j}\\ s_j=\sqrt{\underset{i=1}{\overset{M}{\Σ}}{(x_{i,j}-{\stackrel{\‾}{x}}_j)}^2/(M-1)};\end{array}---\left(5\right)$

Wherein: M is the sample total after combination in matrix.

(4.2.2) offset minimum binary is utilized to extract latent variable group T:

T＝X ^LR

；(6)

R＝W(P ^TW) ^-1

Wherein, T is the latent variable matrix be made up of multiple offset minimum binary latent variable, and R is the matrix of coefficients of deflected secondary air, and W is the weight matrix obtained by offset minimum binary, and P is the load matrix corresponding to T.

(4.2.3) utilize canonical correlation analysis to carry out aftertreatment to offset minimum binary latent variable, obtain final latent variable u:

u＝Tv；(7)

Wherein, v is the weight vectors corresponding to T.Because y is that single argument exports response, due to type correlation analysis method itself, therefore final need be extracted a latent variable.

It is pointed out that the weight matrix of offset minimum binary and type correlation analysis method can be tried to achieve by asking for the proper vector of particular matrix, is the statistical analysis technique of correlationship between a kind of effective analysis data variable.

(4.2.4) low frequency autoregressive model is set up:

The regression coefficient q between latent variable and response variable is asked for by least square method:

q＝(u ^Tu) ^-1u ^Ty ^L；(8)

Therefore based on the autoregressive model of latent variable method establishment be finally:

$\begin{array}{c}{\hat{y}}^L=X^L{\mathrm{\σ}}^L\\ {\mathrm{\σ}}^L=Rvq\end{array};---\left(9\right)$

Wherein, σ ^lthe forecast model regression coefficient obtained by offset minimum binary and canonical correlation analysis two kinds of methods combining, for the low frequency blood glucose prediction value obtained.

After obtaining the forecast model set up for different object, need to verify the versatility of forecast model.Namely the checking of model commonality verifies that the potential blood sugar time-series dynamics of whether Different Individual has similar autocorrelation; Whether the model set up for any individual may be used on other individualities carry out real time blood sugar prediction.Therefore the precision of prediction of contrast two kinds of models is needed for any one object, a kind of is the model (same individual model) utilizing same individual information to set up, and a kind of is the model (Different Individual model) utilizing other individual information to set up.Here, we consider two kinds of same individual models, and the first is the individual forecast model being separated the low frequency blood glucose signal foundation obtained based on frequency band, and the second is the individual forecast model set up based on original or full range blood glucose measurement signal.The contrast of these two kinds of same individual models will illustrate whether the removal of high-frequency signal can impact prediction precision.

Wherein, as follows based on prediction modeling concrete steps that are original or full range blood glucose signal:

For original blood glucose signal, be separated without the need to the frequency band carrying out step 2, directly perform step 3 and 4.Therefore, what modeling here utilized is the blood glucose signal of full range.

Therefore the autoregressive model finally adopting least square method to set up based on original or full range blood glucose signal is:

$\hat{y}=X\mathrm{\θ};---\left(10\right)$

Wherein, θ is the forecast model regression coefficient obtained by offset minimum binary and canonical correlation analysis two kinds of methods combining, for the blood glucose prediction value obtained.

In like manner, based on autoregressive model that is original or full range blood glucose signal employing latent variable method establishment be finally:

$\begin{array}{c}\hat{y}=X\mathrm{\σ}\\ \mathrm{\σ}=Rvq\end{array};---\left(11\right)$

Wherein, σ is the forecast model regression coefficient obtained by offset minimum binary and canonical correlation analysis two kinds of methods combining, for the blood glucose prediction value obtained.

Show the research of individual body Model, when PL<7, precision of prediction increases with PL and increases; And along with the increase of PH, the precision of prediction of model presents downtrending, when PH generally chooses 30 ~ 60min, following blood glucose prediction is comparatively meaningful, can ensure certain precision of prediction and reliability.We verify the versatility of model for PL=7 and PH=30min below.Concrete steps are as follows:

(I) same individual model is utilized to carry out blood glucose prediction

A (), to each object, the frequency band separation method in employing step 2, to after blood glucose signal process, calls the first same individual model and carries out blood glucose prediction.

First, call the low frequency autoregressive model set up based on least square and carry out blood glucose prediction, obtain predicted value wherein, θ ^lfor above according to the regression coefficient vector that training data is obtained.

Then, call the low frequency autoregressive model set up based on latent variable and carry out blood glucose prediction, obtain predicted value wherein, σ ^lit is the regression coefficient vector that offset minimum binary-Canonical Correlation Analysis is obtained.

B (), to each object, for original blood glucose measurement signal, is called the second same individual model and is carried out blood glucose prediction.

First, call the autoregressive model set up based on least square and carry out blood glucose prediction, obtain predicted value wherein, the regression coefficient vector obtained according to training data before being of θ.

Then, call the autoregressive model set up based on latent variable and carry out blood glucose prediction, obtain predicted value wherein, σ is the regression coefficient vector that offset minimum binary-Canonical Correlation Analysis is obtained.

(II) Different Individual model is utilized to carry out blood glucose prediction

To each object, call Different Individual model and carry out blood glucose prediction.Here we are for each individuality, and 26 Different Individual models can be adopted to carry out blood glucose prediction to it.

First, the low frequency autoregressive model called based on least square carries out blood glucose prediction, obtains predicted value ${\hat{y}}^L=x^{LT}\mathrm{\&theta;}.$

Then, the low frequency autoregressive model called based on latent variable carries out blood glucose prediction, obtains predicted value ${\hat{y}}^L=x^{LT}\mathrm{\&sigma;}.$

(III) precision of prediction of two kinds of models is contrasted

First, after carrying out blood glucose prediction for same individuality by (6.1) and (6.2) described method, RMSE (root-mean-square error) performance index are calculated as follows:

$RMSE=\sqrt{\frac{1}{N}\underset{i\&Element;N}{\&Sigma;}{\left(y\right(i)-{\hat{y}}^{*}(i\left)\right)}^2};---\left(13\right)$

Wherein y (i) represents original blood sugar measured, representing the blood glucose prediction result adopting different model to obtain, can be the low frequency blood glucose prediction value adopting the first same individual model or Different Individual model to try to achieve here or adopt the full range predicted value that the second same individual model is tried to achieve n is sample number.Here it is pointed out that when adopting Different Individual model to carry out blood glucose prediction, the different blood glucose prediction results obtained first are averaged, this mean value to be contrasted with reference quantity as predicting the outcome.Meanwhile, when weighing precision of prediction, need original blood sugar measured as reference amount, predicated error is blood glucose prediction result and original blood sugar measured contrast.

Then, based on RMSE index, contrast with the precision of prediction of t hypothesis testing method to two models (same individual model and Different Individual model) of pairing.

Result (as shown in table 1) shows, adopts the autoregressive model prediction precision of latent variable method establishment to be better than least square method.In addition, adopt identical modeling method, the blood glucose prediction result obtained based on two kinds of same individual models to carry out blood glucose prediction based on Different Individual model and have and add up similar precision of prediction.Therefore, all may be used on other individualities carry out real-time blood glucose prediction to any one individual model set up, and latent variable and least square method can not affect versatility.Be separated based on frequency band the low frequency Self-regression Forecast Model set up and be general low frequency model.In addition, since the precision of prediction adopting two kinds of same individual models to obtain is similar, illustrate that the high-frequency signal under the certain cutoff frequency of removal does not affect blood glucose prediction.

The blood glucose prediction Comparative result (RMSE (mg/dL) (MEAN ± MAD) that table 1 adopts different least square/latent variable model to obtain for group 1 (12 object) with group 2 (14 object) two groups individuality.MEAN is mean value, and MAD is mean absolute deviation.)

In addition, show the research of individual body Model, P (threshold value that frequency band is separated), PL (predictive variable length), choosing of PH (forecast interval) all have impact to the precision of model.Present invention incorporates two groups of individual informations to analyze and research to choosing of P, PL, PH.

A () P chooses: first, and Fig. 2 shows the design sketch after adopting frequency band to be separated for P=60min.As seen from the figure, low-frequency data can reflect overall variation tendency and curve is more level and smooth, and this has also embodied and has adopted frequency band separation method can remove the certain noise information of blood glucose signal.Secondly, we analyze the impact chosen model accuracy of P.Use r _m,nrepresent that the general low frequency model set up according to object n observation data carries out the RMSE value of object m blood glucose prediction.The RMSE average r of object m _mcalculated by the result of other 25 objects average and obtain: these calculating repeat under different threshold period P.The r of tested object m _mvalue and threshold period can be arranged in a vector, because the RMSE value of different object has different scopes, r _mvalue is standardized the percentage that is converted between 0 to 100% to draw.The r that object m is converted _mstandard value be defined as: 26 individualities of group 1,2 are as the tested object of general low frequency model.At scope 20min<P<80min, from the t test of hypothesis (α=0.05) of matching, the accuracy Bu Yin Teding P value of general low frequency model and statistically superior to some extent.And along with P continuation increase, significantly reducing appears in the accuracy of universal model.

B () PL chooses: for every group objects, and prediction length PL is evaluated by all 26 objects the impact of certain PH (here for 60min) accuracy of forecast.First determine the general low frequency model of each object, then the test model using its mean value as each object, be applied to the 60min blood glucose prediction of the different object of this group.The mean value (MEAN) of RMSE index and mean absolute deviation (MAD) on average obtain by often organizing tested object.Calculate the RMSE that two groups of tested objects under different PL predict the outcome, as shown in Figure 3 and Figure 4, the precision of prediction of model increases with PL and increases until PL>7 its result.

C () PH chooses: similar with choosing of PL, and PH is evaluated by all 26 objects for the impact of certain PL (here for PL=7) accuracy of forecast.First determine the general low frequency model of each object, then the test model using its mean value as each object, be applied to the blood glucose prediction of the different object of this group.The mean value of RMSE index and intermediate value mean deviation on average obtain by often organizing tested object.Calculate the RMSE that two groups of tested objects under different PH predict the outcome, as shown in Figure 5 and Figure 6, model prediction accuracy declines with the increase of PH its result.Therefore PH when generally choosing 30 ~ 60min following blood glucose prediction comparatively meaningful, can ensure that the enough outside input action time is to eliminate following abnormal plasma glucose on the one hand, certain precision of prediction and reliability can be ensured on the other hand.

In order to the general low frequency model precision of prediction that clearer display the method for the invention is set up, the predicted value of itself and SD model and true measurement compare, as shown in Fig. 7, Fig. 8, Fig. 9 and Figure 10 by we.

Step 5: the on-line prediction based on general low frequency model:

Based on aforementioned 4 steps, we can call the set up future blood glucose value of general predictive model to any individual and predict.Come by following steps:

(5.1), during on-line prediction, any individual new data is being collected afterwards (subscript n ew represents new samples, J=PL), right the frequency band separating treatment of carrying out as described in step 2 obtains

(5.2) call general low frequency model and carry out on-line prediction:

(5.2.1) call the general low frequency model set up based on least square and carry out on-line prediction:

For each new test data the predicted value after PH sampled point is generated by following steps

${\hat{y}}_{new}^L=x_{new}^{LT}{\mathrm{\&theta;}}^L;---\left(14\right)$

Wherein, θ ^lfor above according to the regression coefficient vector that training data is obtained.

(5.2.2) call the general low frequency model set up based on latent variable and carry out on-line prediction:

For each new test data the predicted value after obtaining PH sampled point is calculated by following steps

${\hat{y}}_{new}^L=x_{new}^{LT}{\mathrm{\&sigma;}}^L;---\left(15\right)$

Wherein, σ ^lit is the regression coefficient vector that offset minimum binary-Canonical Correlation Analysis is obtained.

(5.3) precision of prediction is weighed

The prediction of quality result of acquisition and actual measured value are contrasted.After a series of new prediction signal of acquisition, can according to the blood glucose prediction precision of RMSE (root-mean-square error) performance Index Calculation of definition in formula (13) for new samples.

Claims (3)

1. be separated and the general blood glucose prediction method of data modeling based on frequency band, it is characterized in that, the method comprises the following steps:

Step 1: modeling blood glucose signal pre-service: the subcutaneous blood glucose signal of individuality obtained with certain sampling period Δ t is combined into one dimension time series data x ^t(1 × Z), wherein, x is the measured value of blood glucose signal, and Z is number of samples, removes spike noise wherein; Timing dependence and the dynamic-change information of blood glucose signal is contained in this one dimension time series data;

Step 2: blood glucose signal frequency band is separated: the Changing Pattern of dynamic in high and low different frequency range analyzing blood sugar, distinguishes critical frequencies section and secondary frequency band, determines the optimal threshold that frequency band divides; Butterworth LPF is adopted to carry out frequency band separation to blood glucose signal according to the separation threshold value of definition;

Step 3: obtain predictive variable matrix and response matrix: slip over x with the one dimension moving window that a length is K sampled point ^lT(1 × Z), each mobile sampled point, mobile Z-K+1 time altogether; Using the data in each moving window as a new row vector, then can be combined as a two-dimensional data matrix X ^l(N × K), wherein, N=Z-K+1, K=PL+H (PH/5); PL represents the length of predictive variable, and H represents prediction step, and PH represents forecast interval, H with PH linear module is different, but all represent predict be how many walk after blood sugar; Because prediction step is often walk 5 minutes, therefore there is PH=5 × H;

Step 4: the blood glucose prediction modeling be separated based on frequency band: carry out blood glucose prediction modeling by the method based on latent variable;

Step 5: predict according to the future blood glucose value of blood glucose prediction model to any individual that step 4 is set up; Come by following steps:

(5.1), during on-line prediction, any individual new data is being collected after, right the frequency band separating treatment of carrying out as described in step 2 obtains wherein, subscript n ew represents new samples, J=PL;

(5.2) call general low frequency model and carry out on-line prediction:

(5.2.1) call the general low frequency model set up based on least square and carry out on-line prediction: for each new test data the predicted value after PH sampled point is generated by following steps

${\hat{y}}_{new}^L=X_{new}^{LT}{\mathrm{\&theta;}}^L$

Wherein, θ ^lfor above according to the regression coefficient vector that training data is obtained;

(5.2.2) call the general low frequency model set up based on latent variable and carry out on-line prediction:

For each new test data the predicted value after obtaining PH sampled point is calculated by following steps

${\hat{y}}_{new}^L=x_{new}^{LT}{\mathrm{\&sigma;}}^L;$

Wherein, σ ^lit is the regression coefficient vector that offset minimum binary-Canonical Correlation Analysis is obtained;

(5.3) precision of prediction is weighed: the prediction of quality result of acquisition and actual measured value are contrasted; After a series of new prediction signal of acquisition, according to the blood glucose prediction precision of the RMSE performance Index Calculation defined in following formula for new samples, wherein RMSE is root-mean-square error:

$RMSE=\sqrt{\frac{1}{N}\underset{i\&Element;N}{\&Sigma;}{\left(y\right(i)-{\hat{y}}^{*}(i\left)\right)}^2};$

Wherein, y (i) represents original blood sugar measured, represent the blood glucose prediction result adopting different model to obtain, adopt the low frequency blood glucose prediction value that the first same individual model or Different Individual model are tried to achieve or adopt the full range predicted value that the second same individual model is tried to achieve n is sample number.

2. be separated and the general blood glucose prediction method of data modeling based on frequency band according to claim 1, it is characterized in that, described step 2 comprises following sub-step:

(2.1) the single order low pass Butterworth filter that form of choosing is following:

$\tilde{x}\left(k\right)={\mathrm{\&beta;}}_{1}x\left(k\right)+{\mathrm{\&beta;}}_{2}x(k-1)-\mathrm{\&alpha;}\tilde{x}\left(k\right);$

Arranging its threshold values cycle is P (min); Wherein, x is the measured value of blood glucose signal, its filter value, α, β ₁and β ₂be filtering parameter, k is sampled point; Therefore filtering output is the linear combination of history filter value, history and current measurement value; According to the understanding of reality to blood sugar dynamic, P value is generally 40min ~ 80min; (2.2) to the data x gathered in step 1 ^t(1 × Z) carries out filtering by single order Butterworth LPF, isolates low frequency blood glucose level data x ^lT(1 × Z) and high frequency blood glucose level data x ^hT(1 × Z); High frequency blood glucose level data is considered to noise, and remove high frequency blood glucose level data, step is:

x ^LT(1×Z)＝x ^T(1×Z)-x ^HT(1×Z)；

Wherein, subscript L, H represents low frequency and high-frequency data respectively.

3. be separated and the general blood glucose prediction method of data modeling based on frequency band according to claim 1, it is characterized in that, described step 4 adopts the method based on latent variable to carry out blood glucose prediction modeling; Concrete steps are as follows:

(4.1) predictive variable and output variable is chosen: according to the value of PL, choose x ^l(N × K) front PL arranges as prediction matrix X ^l(N × J), J=PL; Then X ^l(N × K) PL+PH arranges as output variable y ^l(N × 1);

(4.2) the method establishment Self-regression Forecast Model of latent variable is adopted according to low frequency blood glucose signal: the autoregression low frequency forecast model based on latent variable adopts the feature extracting method of offset minimum binary-canonical correlation analysis to carry out modeling; First extract preliminary latent variable group with offset minimum binary, the latent variable extracted due to deflected secondary air can not ensure the closely related relation of itself and response variable, with canonical correlation analysis, aftertreatment is carried out to it again, thus extract wherein with relevant variable closely-related part carry out regression modeling and prediction; Step is as follows:

(4.2.1) data prediction

The predictive variable matrix of all individualities and response matrix are grouped together, for the variable x of any point in the predictive variable matrix after combination and response matrix _i,j, average, global criteria process divided by standard deviation are subtracted to this variable, wherein, subscript i representative batch, j represents variable, and the computing formula of standardization is as follows:

$x_{i,j}=\frac{x_{i,j}-{\stackrel{\&OverBar;}{x}}_j}{s_j};$

Wherein: the average of the arbitrary row of matrix after combination, s _jit is the standard deviation of respective column; Its computing formula is as follows:

${\stackrel{\&OverBar;}{x}}_j=\frac{1}{M}\underset{i=1}{\overset{M}{\&Sigma;}}{x}_{i,j}$

$s_j=\sqrt{\underset{i=1}{\overset{M}{\&Sigma;}}{(x_{i,j}-{\stackrel{\&OverBar;}{x}}_j)}^2/(M-1)};$

Wherein: M is the sample total after combination in matrix;

(4.2.2) offset minimum binary is utilized to extract latent variable group T:

T＝X ^LR

；

R＝W(P ^TW) ^-1

Wherein, T is the latent variable matrix be made up of multiple offset minimum binary latent variable, and R is the matrix of coefficients of deflected secondary air, and W is the weight matrix obtained by offset minimum binary, and P is the load matrix corresponding to T; (4.2.3) utilize canonical correlation analysis to carry out aftertreatment to offset minimum binary latent variable, obtain final latent variable u:

u＝Tv；

Wherein, v is the weight vectors corresponding to T; Because y is that single argument exports response, due to Canonical Correlation Analysis itself, therefore final need be extracted a latent variable;

(4.2.4) low frequency autoregressive model is set up:

The regression coefficient q between latent variable and response variable is asked for by least square method:

q＝(u ^Tu) ^-1u ^Ty ^L；

Therefore based on the autoregressive model of latent variable method establishment be finally:

；

σ ^L＝Rvq

Wherein, σ ^lthe forecast model regression coefficient obtained by offset minimum binary and canonical correlation analysis two kinds of methods combining, for the low frequency blood glucose prediction value obtained.