CN105092521B - Milk condensing process method of real-time based on increment principal component analysis - Google Patents
Milk condensing process method of real-time based on increment principal component analysis Download PDFInfo
- Publication number
- CN105092521B CN105092521B CN201510497332.XA CN201510497332A CN105092521B CN 105092521 B CN105092521 B CN 105092521B CN 201510497332 A CN201510497332 A CN 201510497332A CN 105092521 B CN105092521 B CN 105092521B
- Authority
- CN
- China
- Prior art keywords
- msub
- mtr
- mtd
- data
- window
- 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
Links
Abstract
The invention discloses a kind of milk condensing process method of real-time based on increment principal component analysis, on the basis of it is near infrared spectrum data in milk condensing process, using clustering method by data clusters into multiple windows, then dimensionality reduction is carried out to data in window using increment principal component analysis method, and when follow-up lot data is added in window, original data can be corrected by the sample data newly increased, so as to carry out lasting renewal to the characteristic value in window and characteristic vector, characteristic in final dynamic window will accurately reflect milk condensing process.The method of the present invention can carry out more structurally sound data analysis near infrared spectrum data, finally realize the accurate monitoring to milk condensing process.By validation trial, for method of the invention when being detected to milk condensing process, accuracy rate is up to more than 90%, so as to the production for higher quality cheese product.
Description
Technical field
It is more particularly to a kind of to be based on increment principal component analysis the present invention relates to a kind of milk condensing process method of real-time
Milk condensing process method of real-time.
Background technology
The step of milk condensing is used for and makes cheese is to form curdling glue by cutting, and the time cut finally influences milk
The yield and quality of junket.If the time of cutting early causes very much the loss of yield, the time of cutting causes to obtain high-moisture too late
Cheese.During present cheese dairy products make, plant operator is mainly according to their working experience hand cut milk
The curdling glue formed is condensed, not only cost height also carries subjective factor, which prevent the automation of flow and have impact on work effect
Rate.Therefore, we are necessary to optimize a kind of automatic mode for determining clipping time.
Milk curdled milk enzyme induction gel is to need to undergo three root phases：
(1) initial enzyme hydrolysis, J- caseins will change its casein micelles, and ultimately form paracasein.
(2) polymerization of paracasein, wherein rate of polymerization depend on the concentration of free paracasein, it is meant that this stage
It is to rely on speed and the degree of J casein hydrolysis；
(3) gel process, polymer network polymeric micelle chain, also referred to as gel solidification are formed.Stage between excessively, no
Easily discover, because head and afterbody continuous stage are overlapping to a certain extent.
The content of the invention
It is an object of the present invention to provide a kind of milk condensing process based on increment principal component analysis side of monitoring in real time
Method.It can accurately detect the change of main component in milk condensing process, so as to realize the real-time monitoring of milk condensing process,
For the time standard as cutting curdling glue, and then improve the yield and quality of cheese.
Technical scheme：Milk condensing process method of real-time based on increment principal component analysis, including with
Lower step：
S1. the near infrared spectrum data of Each point in time in the milk condensing process of multiple batches is gathered as original number
According to；
S2. the pretreatment being standardized to initial data, the standardized data of multiple batches is obtained；
S3. the standardized data of first batch is carried out by cluster operation by clustering algorithm；, multiple clusters are generated, and
These clusters are identified as different windows；
S4. principal component analysis (conventional principal component analysis method) is carried out to the standardization initial data in each window, obtained
To main composition score corresponding to the data load and data in each window；
S5. according to the corresponding relation at time point, the standardized data of next batch will be increased in each window, and
Using increment principal component analysis method the newly-increased data in window are modified and predicted with their main composition score, and shape
Into the data characteristics of current window；
S6. to remaining batch standardized data repeat step S5 one by one；
S7. the near infrared spectrum real time data in milk condensing process to be monitored is gathered, and with step S5 identical sides
Method increases near infrared spectrum real time data in window newly, is obtained by the main composition for observing near infrared spectrum data in each window
Divide to judge which stage (go to realize by window and observe) milk condensing is in.
In the above-mentioned milk condensing process method of real-time based on increment principal component analysis, the pre- place of the step S2
Reason uses normalization function, i.e.,
In the foregoing milk condensing process method of real-time based on increment principal component analysis, used in the step S3
Index of the Euclidean distance as cluster.Multidimensional Euclidean distance d calculation formula are
Wherein xi1Represent the i-th dimension coordinate of first point, xi2Represent the i-th dimension coordinate of second point.The 1st dimension sit
Mark, xi2Represent xi2The 2nd dimension coordinate, d represent hyperspace point point between geometric distance.It is foregoing based on increment it is main into
In the milk condensing process method of real-time of part analysis, the data characteristics in the step 7 in each window is included in window
The main composition load of data in the numbers of data, window, the time point corresponding to the score and score of main composition.
In the foregoing milk condensing process method of real-time based on increment principal component analysis, in the step S5, when
There are the data for not meeting the condition for adding existing window in the standardization initial data of next batch, then establish new window,
And these data are put into newly-built window, and the calculating of load and score is carried out using principal component analysis method；If window
Quantity reaches the upper limit, then merges two nearest windows, otherwise directly establish a new window.
Increment principal component analysis method (IPCA) in the step S5, it need not assume input sample for zero-mean and can
To dynamically update average, it is described as follows：
Give d*n near infrared spectrum data matrix Xn={ X1,X2,...Xn, wherein, XiRepresent i-th batch
Data (d represents d dimension vectors).
, can be in the hope of PCA score matrixes F using PCA algorithms for the data of the near infrared spectrum of first batch0：
X0Q0=F0
Wherein P0Represent the load of the 0th lot data, i.e. unit character vector.
OrderK0It is a constant.
As given new input sample X1, have as follows
Wherein Q1Represent the load of the 1st lot data.
Wherein
In summary, for each new input sample Xn, the sample newly inputted can be entered according to sample before
Row amendment, and predict its characteristic value and characteristic vector in whole set.
Compared with prior art, on the basis of near infrared spectrum data of the present invention in milk condensing process, cluster is utilized
Data clusters into multiple windows, are then carried out dimensionality reduction by method using increment principal component analysis method to data in window, and
And when follow-up lot data is added in window, original data can be corrected by the sample data newly increased, so as to
Lasting renewal can be carried out to the characteristic value in window and characteristic vector, the characteristic in final dynamic window will be accurately anti-
Mirror milk condensing process.The method of the present invention combines clustering method and increment principal component analysis method based on window, can
To carry out more structurally sound data analysis near infrared spectrum data, the accurate monitoring to milk condensing process is finally realized.
By validation trial, for method of the invention when being detected to milk condensing process, accuracy rate is up to more than 90%, can
For producing higher quality cheese.
Brief description of the drawings
Fig. 1 is the schematic flow sheet of the present invention；
Fig. 2 is the near infrared spectrum data dimension figure of embodiments of the invention；
Fig. 3 is the first batch PC1 score chart of percentage comparisons of embodiments of the invention；
Fig. 4 is the amendment of second lot data and predicts PC1 scores and unmodified PC1 graphs of a relation；
Fig. 5 is the error rate schematic diagram of the embodiment of the present invention.
Embodiment
With reference to embodiment, the present invention is further illustrated, but is not intended as the foundation limited the present invention.It is right
Any formal accommodation and/or change that the present invention is made fall within the scope of the present invention.
Embodiments of the invention：Milk condensing process method of real-time based on increment principal component analysis, such as Fig. 1 institutes
Show, comprise the following steps：
S1. the near infrared spectrum data of Each point in time in the milk condensing process of multiple batches is gathered as original number
According to；
S2. the pretreatment being standardized to initial data, the standardized data of multiple batches is obtained；
S3. the standardized data of first batch is carried out by cluster operation by clustering algorithm, generates multiple clusters, and will
These clusters are identified as different windows；
S4. principal component analysis is carried out to the standardization initial data in each window, the data obtained in each window carry
Lotus, and calculate main composition score；
S5. according to the corresponding relation at time point, the standardization initial data to next batch is increased into each window
In, and using increment principal component analysis method the newly-increased data in window are modified and predicted they load and it is main into
Part score, and form the data characteristics of current window；
S6. to the standardized data of remaining batch repeat step S5 one by one；
S7. the near infrared spectrum real time data in milk condensing process to be monitored is gathered, and with step S5 identical sides
Near infrared spectrum real time data is referred in respective window by method, the principal component according to the near infrared spectrum real time data in window
Score judges current milk condensing state, and carries out respective handling.e
It can use following steps when being embodied：
The present embodiment uses the near infrared spectrum data of 12 batches of milk condensing process.
Step 1：The extraction of data, the data structure included in each batch are the corresponding time point t of an x.Wherein t
For the time of milk condensing process, x is the near infrared spectrum data of 1557 dimensions at corresponding t time points.
batch1：(x11,t11),(x12,t12)...(x1n,t1n)
batchm：(xm1,tm1),(xm2,tm2)...(xmn,tmn)
Step 2：Data are pre-processed, to batch1Pre-processed to obtain X0, pre-process and use normalization function.
I.e.
Step 3：Calculate X0Multidimensional Euclidean distance d
Wherein xi1Represent the i-th dimension coordinate of first point, xi2Represent the i-th dimension coordinate of second point.
Change into the square formation A on multidimensional Euclidean distance1
Following steps 4-7, window is established by clustering algorithm, and calculate the principal component scores of data.
Step 4：It is A to set initial value i=1, j=1, n=1 (1≤n≤k) wherein k1Total columns.If anj(σ is≤σ
One threshold value of setting), σ ＞ 0, then original (x1n,t1n) it is put into a window windowiIn, if an"j＜ σ then enter step 5, if
an"j＞ σ, then j=n ", n=n ", i=i+1 repeats this step.I represents window counter in step 4, represents i-th of window, and
N and j difference representing matrixs A1The coordinate of middle row and column, a position is determined, takes line position to put and be changed into n ", column position is also changed into n "
a。
Step 5：If n<K, then n=n+1, and return to step 4, enters step 6 if n=k.
Step 6：Data in window carry out principal component analysis, and loading matrix and principal component scores is calculated.
For windowiInterior data X, is handled as follows
X=P Δs QT
F=P Δs=P Δs QT=XQ
Obtain windowiLoad QiWith initial time ti1With deadline ti2、(x1n,t1n) PCA scores F.
Step 7：To the data of second lot, load is predicted using incremental principal component analysis.
For batch2:(x21,t21),(x22,t22)...(x2n,t2n), if ti1≤t2n≤ti2, then by x2nIt is added to window
windowi, and obtained x2nRevised PCA is scored atLoad is predicted using incremental principal component analysis：
Wherein
If ti1＞ t2nOr t2n>ti2, re-creating a new window window, (definable one is different from existing window
Subscript), by data x2nIt is put into the window, is handled as follows：
X=P Δs QT
F=P Δs=P Δs QT=XQ
Obtain the window load Q and data corresponding to PCA score F, Q and F here be identical with the small tenon of window.
(in step 7, t2nThe time is represented, if t2nIn time ti1And ti2If window is just added between not in ti1And ti2
Between just re-create a window).
Step 8：For multiple batches of follow-up incoming data, for batchn:(xn1,tn1),(xn2,tn2)...(xnn,tnn)
The method of repeat step 7.
Step 9：After the completion of the data processing of all batches, freshly harvested nearly red line spectrum real time data is used and step
Rapid 7 identical method be referred in existing window, and the main composition of the nearly red line spectrum real time data in prediction window obtains
Point, concrete operation formula is as follows
Wherein
The data newly increased may finally in real time be monitored by above-mentioned algorithm, observed using multiple windows each
Some features of data in window, such as characteristic vector, PCA load and PCA scores, and with incremental principal component analysis to each entrance
Newly-increased data in window are modified and predicted their load and PCA scores.
Near infrared spectrum data is measured by near-infrared spectrometers and SMA fiber reflection probe, in the present embodiment
Using 12 batch datas, the form per batch data is a time point corresponding 1557 dimension near infrared spectrum data, and following table is first
The data sample of secondary cheese condensation process.
If for the t=0 of first batch, t=4.8, t=6.7, the near infrared spectrum data of t=12.2 minutes is made
2-D data is illustrated, and two dimensional image as shown in Figure 2 can be formed, more intuitively to analyze its feature.
The overall process of milk condensing can be clearly found out from figure.Abscissa is 0-1557 dimensions, and ordinate is near red
Near infrared spectrum data measured by external spectrum analyzer.During showing milk condensing in figure, three phases are experienced
Change, causes main composition constantly changing.
Using 12 lot data samples about 2/3 of milk condensing process as training data, principal component pca model is established,
1/3 data are as inspection data.
The first batch data X of 1557 dimensions is decomposed first, the apposition sum of m vector is expressed as, is expressed as
Meet f in formulai=Xqi, i=1,2 ..., m.When X is present it is a certain degree of linearly related when, then X information content master
Before being embodied on q load vectors, therefore have
Above q pivot represents the main information of data, and E represents error matrix, can be ignored, data X is used
[f1,f2,...fq] approximate representation, i.e.,
Q ＜ ＜ m, it is achieved thereby that the compression of data.The key of PCA algorithms is to solve for load vectors, conventional method master
If singular value decomposition and Eigenvalues Decomposition etc..The method that method used herein is mainly based upon singular value decomposition.
Sample data matrix X singular value decomposition school timetable is shown as
X=U Ε VT
U=[u in formula1u2…un]∈Rn×n, V=[v1v2…vm]∈Rm×m
Wherein UTU=VTV=I, π1＞ π2＞ ... ＞ πmFor matrix X singular value.And then：
WhereinIt is Σ characteristic value,And viIt is Σ characteristic vector, therefore has Xvi=piui
Make piui=fi,qi=vi, so having
PCA dimensionality reduction numbers are carried out to 12 batch datas by above-mentioned singular value decomposition method, and calculate the explanation of cumulant variable
Degree.The information summation that sample data is included is λ1+λ2+…+λm, the information included in i-th of pivot is λi, it is to data
Explanation rate be λi/(λ1+λ2+…+λm).The explanation rate of general pivot is typically chosen as 85%, be the advantages of this method it is simple,
Amount of calculation is smaller, reliable.The data of 4 batchings time explanation degree of cumulant variable in principal component 1 and principal component 2 all account for 90% with
On, principal component 1 is far longer than principal component 2.Replacement of the principal component 1 as initial data is used herein, hereinafter will no longer be illustrated.
PCA dimension-reduction treatment is carried out to the near infrared spectrum data of 12 batches by the method for singular value decomposition, it is final to obtain
To the PC1 scores corresponding to each batch Each point in time.From figure 3, it can be seen that milk condensing process Main change trend,
It is to belong to which of milk condensing process stage that we, which can not accurately estimate these wave bands, i.e., initial enzyme hydrolysis, secondary
The polymerization of casein, gel process.But we can make function of the PC1 scores as the time, so as to further observe
The change of data.
We can only hydrolyze at 7~9 minutes initial enzyme in rough observing since Fig. 3, pair at 10~45 minutes
The polymerization of casein and gel process.
By a series of data processing, we pass through each window by a batch of data clusters to 17 windows
Mouthful, it is observed that some parameters of each window, such as the load vectors of each window, the number for the data that each window is included
The initial time point and end time point of amount and each window.
Some parameters such as following table in window.
This shows the near infrared spectrum data of first batch after window, and divide into 17 classes has the number of identical characteristic
According to the data in window 1, window 2, window 3 are most, it is possible thereby to judge, window 1 is that initial enzyme starts hydrolysis stage, window
The polymerization stage of paracasein when mouth 2 is, window 3 is the gel process stage.Said process and result are just for independent batch
The processing of progress, the near infrared spectrum data of independent batch can be observed, and the change of real-time monitoring data, but deposit
In some errors.
After the data of second lot to be introduced to the window of first batch data, by the processing of incremental principal component analysis, the is obtained
The revised PC1 scores of two lot datas.Fig. 4 shows the amendment of second lot data and predicts PC1 scores and not
PC1 relation is corrected, the PC1 scores as can be seen from the figure predicted are more or less the same with original PC1 scores, and have trickle change
Change.It is that first data have carried out trickle amendment to the data of second lot the reason for the above results occur, and to original
Carry out data result and do not cause huge deviation.
Herein by R2The above results are tested,Here,
Its normal span is 0 to 1, closer to 1, shows that the variable of equation is stronger to y interpretability, this model
To the more accurate of spectroscopic data condition adjudgement, R is calculated in this experiment2Value be 0.98, illustrate that the present invention can be by right
The cluster of ir data and increment principal component analysis, it can be good at judging that milk is specific during cheese is condensed into
State of aggregation.
As can be seen from Figure 4 28 minutes and 66 minutes 2 time points are there is huge deviation, should reject during inspection different
Regular data so that assay is more accurate.Error rate be can be seen that in 5% within 0~60 minute from Fig. 5, individual data exists
10% or so.Illustrate that the coagulated state that the algorithm can predict that newly-increased data are stated is feasible, method of the invention can utilize
Existing data are modified to newly-increased data, and error can be controlled within 10%.
Claims (5)
1. the milk condensing process method of real-time based on increment principal component analysis, it is characterised in that comprise the following steps：
S1. the near infrared spectrum data of Each point in time in the milk condensing process of multiple batches is gathered as initial data；
S2. pretreatment is standardized to initial data, obtains the standardized data of multiple batches；
S3. the standardized data of first batch is carried out by cluster operation by clustering algorithm, generates multiple clusters, and by these
Cluster is identified as different windows；
S4. principal component analysis is carried out to the standardized data in each window, obtains the data load in each window, and calculate
Go out main composition score；
S5. according to the corresponding relation at time point, the standardization initial data to next batch is increased in each window, and
It is modified and predicts that their load and main composition obtain to the newly-increased data in window using increment principal component analysis method
Point, the data characteristics of current window is formed, the specific method for predicting load and main composition score is to use formula
<mrow>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<msub>
<mi>F</mi>
<mn>1</mn>
</msub>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
<mrow>
<msub>
<mi>Q</mi>
<mn>1</mn>
</msub>
<mo>=</mo>
<msup>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>F</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>F</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>=</mo>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<msup>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mi>T</mi>
</msup>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
<mo>)</mo>
</mrow>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<msup>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mi>T</mi>
</msup>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>F</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>F</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>=</mo>
<msubsup>
<mi>K</mi>
<mn>1</mn>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<msup>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mi>T</mi>
</msup>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>F</mi>
<mn>0</mn>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>F</mi>
<mn>1</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
Wherein X0Represented for the matrix of original standardized data in window, X1For the standardized data of increased next batch
Matrix expression, F0And Q0For X0Principal component scores and the expression of the matrix of load, F1And Q1Principal component scores and load for prediction
Matrix represents；
S6. to the standardized data of remaining batch repeat step S5 one by one；
S7. gather the near infrared spectrum real time data in milk condensing process to be monitored, and with step S5 identical methods
Near infrared spectrum real time data is referred in respective window, the principal component according to the near infrared spectrum real time data in window obtains
Divide to judge current milk condensing state, and carry out respective handling.
2. the milk condensing process method of real-time according to claim 1 based on increment principal component analysis, its feature
It is：The pretreatment of the step S2 uses normalization function.
3. the milk condensing process method of real-time according to claim 1 based on increment principal component analysis, its feature
It is：Using index of the Euclidean distance as cluster in the step S3.
4. the milk condensing process method of real-time according to claim 1 based on increment principal component analysis, its feature
It is：Data characteristics in the step 7 in each window includes the number of data in window, the main composition of data carries in window
Lotus, main composition score and score corresponding to time point.
5. the milk condensing process method of real-time according to claim 1 based on increment principal component analysis, its feature
It is：In the step S5, when not meeting the condition for adding existing window in the standardization initial data of next batch, build
Vertical new window, and place data into newly-built window, and carry out load and principal component scores using principal component analysis method
Calculate；If number of windows reaches the upper limit, merge two nearest windows, otherwise directly establish a new window.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510497332.XA CN105092521B (en) | 2015-08-13 | 2015-08-13 | Milk condensing process method of real-time based on increment principal component analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510497332.XA CN105092521B (en) | 2015-08-13 | 2015-08-13 | Milk condensing process method of real-time based on increment principal component analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105092521A CN105092521A (en) | 2015-11-25 |
| CN105092521B true CN105092521B (en) | 2017-11-14 |
Family
ID=54573437
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510497332.XA Active CN105092521B (en) | 2015-08-13 | 2015-08-13 | Milk condensing process method of real-time based on increment principal component analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105092521B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104834938A (en) * | 2015-04-30 | 2015-08-12 | 北京环境特性研究所 | Hyper-spectral information extraction method based on main component and cluster analysis |
| CN104807774B (en) * | 2014-01-29 | 2017-05-03 | 中国城市规划设计研究院 | Near infrared spectroscopy analysis method of liquid sample |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5206335B2 (en) * | 2008-11-07 | 2013-06-12 | アステラス製薬株式会社 | Principal component analysis method, principal component analysis apparatus, heterogeneous product detection apparatus, principal component analysis program, and recording medium on which principal component analysis program is recorded |
-
2015
- 2015-08-13 CN CN201510497332.XA patent/CN105092521B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104807774B (en) * | 2014-01-29 | 2017-05-03 | 中国城市规划设计研究院 | Near infrared spectroscopy analysis method of liquid sample |
| CN104834938A (en) * | 2015-04-30 | 2015-08-12 | 北京环境特性研究所 | Hyper-spectral information extraction method based on main component and cluster analysis |
Non-Patent Citations (6)
| Title |
|---|
| 《Monitoring Quality Loss of Pasteurized Skim Milk Using Visible and Short Wavelength Near-Infrared Spectroscopy and Multivariate Analysis》;H. M. Al-Qadiri等;《Journal of Dairy Science》;20081231;第91卷(第3期);第950-958页 * |
| 《SELECTION OF NEAR-INFRARED WAVELENGTHS FOR MONITORING MILK COAGULATION USING PRINCIPAL COMPONENT ANALYSIS》;D.Saputra等;《TRANSACTIONS OF THE ASAE》;19921231;第35卷;第1597-1605页 * |
| 《The potential effect of FOS and inulin upon probiotic bacterium performance in curdled milk matrices》;Dina Rodrigues等;《LWT - Food Science and Technology》;20111231;第44卷;第100-108页 * |
| 《基于主成分分析法的奶粉特性指标分类》;赵笠彤;《食品科技》;20131231;第38卷(第6期);第303-306页 * |
| 《基于增量主成分分析和共生梯度方向直方图的人体检测》;潘静等;《天津职业技术师范大学学报》;20110331;第21卷(第1期);第22-24页 * |
| 《荧光光谱成像技术结合聚类分析及主成分分析的藻类鉴别研究》;梁曼等;《光谱学与光谱分析》;20140831;第34卷(第8期);第2132-2136页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105092521A (en) | 2015-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chakraborty et al. | Applications of WASPAS method as a multi-criteria decision-making tool | |
| CN107451101B (en) | Method for predicting concentration of butane at bottom of debutanizer by hierarchical integrated Gaussian process regression soft measurement modeling | |
| CN101710235B (en) | Method for automatically identifying and monitoring on-line machined workpieces of numerical control machine tool | |
| CN107220734A (en) | CNC Lathe Turning process Energy Consumption Prediction System based on decision tree | |
| CN105334185B (en) | The near-infrared model maintaining method differentiated based on spectrum projection | |
| TW201615844A (en) | Method and system of cause analysis and correction for manufacturing data | |
| WO2021036546A1 (en) | Near-infrared quantitative analysis model construction method based on biased estimation | |
| CN107544253A (en) | Based on the retired method of controlling security of large-scale missile equipment for improving Based on Entropy method | |
| CN106371427A (en) | Industrial process fault classification method based on analytic hierarchy process and fuzzy fusion | |
| CN105574587B (en) | A kind of online operating mode course monitoring method of plastic injection molding process | |
| CN108920863A (en) | A kind of method for building up of robot servo system energy consumption estimation model | |
| CN102601881B (en) | Method for monitoring on-line quality and updating prediction model of rubber hardness | |
| CN112904810B (en) | Process industry nonlinear process monitoring method based on effective feature selection | |
| WO2016004749A1 (en) | Method for recognizing tool abrasion degree of large numerical control milling machine | |
| CN108090628A (en) | A kind of grain feelings security detection and analysis method based on PSO-LSSVM algorithms | |
| CN108399255A (en) | A kind of input data processing method and device of Classification Data Mining model | |
| CN104634265B (en) | A kind of mineral floating froth bed soft measurement method of thickness based on multiplex images Fusion Features | |
| CN114898220B (en) | Intelligent production control method for structural member of overhead working truck | |
| CN108133090A (en) | A kind of high-end complex equipment analysis method for reliability of reliability susceptibility driving | |
| CN114754973A (en) | Wind tunnel force measurement test data intelligent diagnosis and analysis method based on machine learning | |
| CN105092521B (en) | Milk condensing process method of real-time based on increment principal component analysis | |
| CN106485049B (en) | A kind of detection method of the NIRS exceptional sample based on Monte Carlo cross validation | |
| US20220277330A1 (en) | Method and system for product processing price prediction based on multiple regression model | |
| CN111932081A (en) | Method and system for evaluating running state of power information system | |
| CN116414095A (en) | Data-driven optimization method for technological parameters in traditional Chinese medicine manufacturing process |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |