CN101145030B - Method and system for increasing variable amount, obtaining rest variable, dimensionality appreciation and variable screening - Google Patents
Method and system for increasing variable amount, obtaining rest variable, dimensionality appreciation and variable screening Download PDFInfo
- Publication number
- CN101145030B CN101145030B CN200610154140XA CN200610154140A CN101145030B CN 101145030 B CN101145030 B CN 101145030B CN 200610154140X A CN200610154140X A CN 200610154140XA CN 200610154140 A CN200610154140 A CN 200610154140A CN 101145030 B CN101145030 B CN 101145030B
- Authority
- CN
- China
- Prior art keywords
- variable
- module
- appreciation
- screening
- those
- 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.)
- Expired - Fee Related
Links
Images
Abstract
The present invention relates to a method and a system to increase the number of variables, to gain remaining variables and to indentify dimensions as well as to screen variables, the method of the present invention can be used to overcome the defect that to gain the system degree of freedom and to careen the system variables need to depend on artificial experience during building up models by the well-known technology; the present invention provides a method that the system degree of freedom and careening the system variables can be gained automatically by the means that the principal value analysis can be used for cooperating the method of estimating the target value as well as limiting condition bases; the system degree of freedom and system variables gained by the present invention can be used for building up an accurate system model; the present invention can be applied to a computer and can provide a variable function of model building for a electrical device linked outside, and can manipulate the computer precisely according to the gained system model.
Description
Technical field
The present invention is the appreciation of a kind of system model dimension and significant variable method for screening and system, refers to a kind of appreciation of system model dimension and significant variable method for screening and system when being used for model construction especially.
Background technology
When variable analysis and system model construction, very important of two factors arranged: supposition and system's independence independent variable of degree of freedom in system (degree of freedom), that is the screening of significant variable.In real system, the number of variable is often very numerous and diverse numerous, add the expansion of time series data item, increase is set up the degree of difficulty of system model, with the empirical mode is example, too much input variable will cause the requirement of data data volume is risen, increased Time Created, improves resource that pattern complexity and waste calculate or the like problem, and and then have influence on the performance of one-piece pattern prediction.Yet in fact, variable in system model, the dependence relation that all has linear/non-linear to a certain degree to each other, real independently independent variable exists only in sub-fraction wherein, so before carrying out system model foundation or error analysis, if can carry out variable analysis and judgement at data earlier, then can effectively reduce because of importing the problem that multivariate causes.On the step of carrying out variable analysis and judgement, at first, the degree of freedom (perhaps can be referred to as its independent dimension (independent dimensions)) that must inquire into system earlier is to understand the number quantity of independent variable, then from numerous variablees, select the appropriate ones that is no less than this number again, under the prerequisite of excessive distortion not, reduce total variable dimension, so as to simplifying its problem to be processed.
There is dual mode can obtain the supposition of above-mentioned degree of freedom in system and the screening of system's independence independent variable in the industry cycle, first mode, be to cooperate with slip-stick artist with professional experiences or operating personnel, the formed knowledge base of experience of accumulating for many years by slip-stick artist or operating personnel, various relevant information of the system that gives and suggestion obtain degree of freedom in system and system's independence independent variable.Yet the knowledge that most professional had is the supposition that obtains via experience, might not and go deep into comprehensively, also sufficient calculation and analysis may not be arranged, so resulting result usually varies with each individual.
Second mode, then be to rely on some basic statistics skills, as STUDY ON PRINCIPAL VALUE method (PCA), come degree of freedom in system is done supposition, and choose suitable variable via comparison modes such as related coefficients, the benefit of doing like this is can be more extensively and in depth inquire into system, the various relations that situational variables exists each other, but being STUDY ON PRINCIPAL VALUE method itself, problem do not provide any clear and definite degree of freedom in system and significant variable information, the user still needs to be made a determination voluntarily by the result of STUDY ON PRINCIPAL VALUE method, generally speaking, coming the Common Criteria of decision-making system degree of freedom by the result of STUDY ON PRINCIPAL VALUE method is by explaining the minimum threshold value of degree from ordering according to system performance and self experience from ordering an eigenwert and systematic variation by the user, the minimum main value number that satisfies this threshold value is the degree of freedom in system of supposition, Common Criteria is done the choice variable conclusion, yet the characteristic of the setting of threshold value and system itself has very confidential relation in this Common Criteria, there is different settings in different systems, except relying on self experience or by the trial and error pricing (trial and error), do not have other relatively clearer and more definite effectively way, in addition the STUDY ON PRINCIPAL VALUE method does not also have at the screening of significant variable clearly provides any effective computational logic, therefore for general user, the STUDY ON PRINCIPAL VALUE method is also inconvenient on using, be difficult to the simple conclusion that relies on its making decision property, in other words, the result of STUDY ON PRINCIPAL VALUE method is relevant with applied object mostly, the STUDY ON PRINCIPAL VALUE method is last only can to obtain a stack features value and a proper vector, but how therefrom decision systems degree of freedom and decision variable object then still depend on experience and systematic knowledge, the knowledge frequently that must have two aspects when therefore using simultaneously concurrently is made judgement, and its result also may vary with each individual.In fact, must be under most situation according to field and the object used, in conjunction with relative professional, the mode that gives statistical study personnel suggestion is analyzed.
The case exposure utilized the STUDY ON PRINCIPAL VALUE method to carry out the application in various fields before many patents were arranged at present, as TaiWan, China patent announcement number I230263 number " based on too development and control purpose in order to quantize the homogeneity type and to include the method for expertise in ", utilize the STUDY ON PRINCIPAL VALUE method collect, demarcate incremental data with unevenness on depicted features and the analyzing semiconductor wafer and feedback is provided and control with the guiding semiconductor fabrication.
Again, as TaiWan, China patent announcement number I235311 number " expert knowledge methods and the system that supply data analysis to use ", utilize the STUDY ON PRINCIPAL VALUE method to produce a model and the multivariate statistics number of new data set, to analyze the good corrupt method of the wafer-process operation of carrying out at wafer-process equipment.
Again,, utilize the STUDY ON PRINCIPAL VALUE method to diagnose the method for a disposal system, especially about the utilization of the PCA that upgrades as TaiWan, China patent announcement number No. 200515112 " utilizing self-adaptation multivariate analysis method to diagnose the method and system of a disposal system ".Please refer to Fig. 1, for during the semiconductor fabrication in order to the monitoring method process flow diagram of the disposal system of treatment substrate, this method comprises the data (S100) that obtain a plurality of observations from the processing system, reach scale-up factor in the utilization static set and set up STUDY ON PRINCIPAL VALUE model (S102), obtain other data (S104) from the processing system, utilize self-adaptation to concentrate and scale-up factor (S106), data and STUDY ON PRINCIPAL VALUE model by other determine at least one statistic (S108), set a control limit (S110), with at least one statistic and control limit make comparisons (S112), detect the error (S114) of disposal system, and notifying operation person (S116).
In above-mentioned those patent documentations, be not difficult to find mainly to be to utilize variable that STUDY ON PRINCIPAL VALUE obtains to produce model, but those patent documentations all clear and definite criterion of neither one are done than the conclusive conclusion of tool, can't determination degree of freedom in system and decision variable object, therefore can't provide model accurately by the produced model of above-mentioned those patent documentations.
Summary of the invention
In view of above problem, the object of the present invention is to provide the appreciation of a kind of system model dimension and significant variable method for screening and system, the method that increases variable quantity, obtains remaining variable is provided, method of the present invention can be found out the desired variable of the accurate degree of freedom in system and the system that filters out, utilize degree of freedom in system and variable can set up out system model accurately, re-use the system model set up out to be applied to a computer, the system of the variable function of a model construction can be provided the electric equipment that the outside links.
To achieve the above object, the present invention proposes appreciation of a kind of system model dimension and significant variable method for screening, comprising: read a plurality of data; Utilize the defined correlated condition of a user that the data that read are carried out a variable classification, this variable classification comprises a plurality of output variables and a plurality of input variable that this user will preset; Utilize at least one proper vector and at least one eigenwert of STUDY ON PRINCIPAL VALUE method in the hope of those input variables; A halt that determines those eigenwerts to calculate is a degree of freedom in system; Obtain first variable of those proper vectors; Stipulate an estimating target rule of those output variables and those input variables; According to this estimating target rule, this first variable and this degree of freedom in system is that the basis is to obtain remaining variable till required variable quantity satisfies; Obtain the variable quantity of an initial proposed; And set up a system model of the variable quantity of this initial proposed.
Rephrase the statement, just, comprise reading of data; Utilize the defined correlated condition of user that the data that read are divided into input variable and output variable; Utilize proper vector and the eigenwert of STUDY ON PRINCIPAL VALUE method in the hope of input variable; The halt that is calculated by eigenwert is with the decision systems degree of freedom; Obtain first variable of proper vector; Stipulate the estimating target rule of output variable and input variable; According to estimating target rule, first variable and degree of freedom in system is that the basis is to obtain remaining variable till variable quantity satisfies; Obtain the variable quantity of initial proposed; And set up the system model of the variable quantity of initial proposed.
Moreover, the present invention also proposes the system of appreciation of a kind of system model dimension and significant variable screening to reach purpose of the present invention, system applies of the present invention is in a computer, and the electric equipment that the outside links is provided the Variables Selection function of a model construction, this system comprises the input data module, and it can obtain data by the Data Source place; Variable quantity module is electrically connected at this input data module, uses for the required maximum variable quantity of this model construction of a user definition; The variable classification module is electrically connected at this variable quantity module, and it can be categorized as at least one input variable and at least one output variable according to the defined correlated condition of this user with those data; The STUDY ON PRINCIPAL VALUE module, it can carry out input variable STUDY ON PRINCIPAL VALUE to obtain proper vector and eigenwert; The degree of freedom in system module, it can be by declining to a great extent the second time in the eigenwert a little as the variable quantity of all eigenwerts before that declines to a great extent some the second time in degree of freedom in system and the acquisition eigenwert; Variable number computing module is electrically connected at this degree of freedom in system module, and it can be caught by the degree of freedom in system module and get variable to be increased; The choice variable module is electrically connected at this variable number computing module, and it can be that the basis obtains choice variable by increasing variable number computing module, the first variable module and estimating target unit; And set up mode module, its choice variable that the choice variable module can be obtained is set up pattern.
Rephrase the statement, just, system applies of the present invention can provide the variable function of a model construction to the electric equipment that the outside links in a computer, and this system comprises: an input data module, and it can obtain a plurality of data by a Data Source place; One variable quantity module is electrically connected at this input data module, uses for the required maximum variable quantity of this model construction of a user definition; One variable classification module is electrically connected at this variable quantity module, and it can be categorized as at least one input variable and at least one output variable according to the defined correlated condition of this user with those data; One STUDY ON PRINCIPAL VALUE module, it can carry out those input variables STUDY ON PRINCIPAL VALUE to obtain at least one proper vector and at least one eigenwert; One degree of freedom in system module, it can be by declining to a great extent the second time in those eigenwerts o'clock as a degree of freedom in system and the variable quantity of all eigenwerts before obtaining second time in those eigenwerts to decline to a great extent a little; One variable number computing module is electrically connected at this degree of freedom in system module, and it can be caught by this degree of freedom in system module and get at least one variable to be increased; One choice variable module is electrically connected at this variable number computing module, and it can be that the basis obtains a plurality of choice variables by this variable number computing module, one first variable module and an estimating target unit; Reach one and set up mode module, its those choice variables that this choice variable module can be obtained are set up pattern.
The present invention also provides the method for the increase variable quantity of appreciation of a kind of system model dimension and significant variable method for screening, and this method comprises:
Catch and get one of this degree of freedom in system and wait to increase variable or increase by a system variable number; Explain that this waits to increase a STUDY ON PRINCIPAL VALUE degree of variation of variable; Compare pre-conditioned whether the meeting that this waits to increase variable and this system; And output meets this pre-conditioned variable to be increased after comparison.
The present invention also provides a kind of method of remaining variable of acquisition of Variables Selection method of model construction, and this method comprises:
Obtain a variable to be increased; Detecting this waits to increase variable and whether meets this estimating target rule; If meet, then this waits to increase variable promptly becomes a suitable variable; If do not meet, then continue to obtain this variable to be increased; And judge whether to reach this variable quantity.
Adopt the present invention can solve the defective that known STUDY ON PRINCIPAL VALUE method can't provide appreciation of conclusive system model dimension and significant variable screening, method of the present invention utilizes the resulting result of STUDY ON PRINCIPAL VALUE method further to analyze, need not possess the user under the situation of sufficient STUDY ON PRINCIPAL VALUE method and system's relevant knowledge and inessential intervention, automatically calculate and search and decline to a great extent for the second time a little to obtain the degree of freedom in system of suggestion, thereafter and according to this result sets up the qualifications storehouse and the estimating target rule of (user's intervention) or this method predetermined (user stays out of) again by the user, Automatic sieve is selected conclusive significant variable, and the appreciation of system model dimension and the significant variable that then can utilize method of the present invention to obtain offer the outside electric equipment that links.
Relevant characteristics and implementation of the present invention, conjunction with figs. is made most preferred embodiment and is described in detail as follows now.
Description of drawings
Fig. 1 is in order to the monitoring method process flow diagram of the disposal system of treatment substrate during the known semiconductor fabrication;
Fig. 2 is the appreciation of system model dimension and the significant variable method for screening process flow diagram of first embodiment of the invention;
Fig. 3 is the method flow diagram of the increase variable quantity of first embodiment of the invention;
Fig. 4 is the method flow diagram of remaining variable of acquisition of first embodiment of the invention;
Fig. 5 is the appreciation of system model dimension and the significant variable method for screening detail flowchart of second embodiment of the invention; And
Fig. 6 is the system schematic of system model dimension of the present invention appreciation and significant variable screening.
The main element description of reference numerals:
The Variables Selection system 1 of model construction
Electric equipment 2
Input data module 10
Variable quantity module 12
Variable classification module 14
STUDY ON PRINCIPAL VALUE module 16
Relevant group module 18
The first variable module 20
Degree of freedom in system module 22
Variable number computing module 24
Qualifications storehouse 240
Explanation degree unit 242
Comparing unit 244
Output variable unit 246
Choice variable module 26
Estimating target unit 260
Choice variable unit 262
Judgment variable module 28
Set up mode module 30
Mode error module 32
Embodiment
Please also refer to Fig. 2, Fig. 3 and Fig. 4, Fig. 2 is the appreciation of system model dimension and the significant variable method for screening process flow diagram of first embodiment of the invention, Fig. 3 is the method flow diagram of the increase variable quantity of first embodiment of the invention, and Fig. 4 is the method flow diagram of the suitable variable of selection of first embodiment of the invention.In Fig. 2, its method flow is to comprise reading a plurality of data (S200) that wherein those data can be obtained by the database that links on the line, a database or any type of file.The user defines the required maximum variable quantity (S202) of this model construction.Utilize the defined correlated condition of user that those data that read are carried out the variable classification (S204) of a plurality of output variables and a plurality of input variables, wherein this variable classification includes but not limited to a flow, a pressure, a temperature or a volume.Utilize at least one proper vector and at least one eigenwert (S206) of STUDY ON PRINCIPAL VALUE method in the hope of those input variables, set a relevant group, to carry out group's classification (S208) via those input variables after this STUDY ON PRINCIPAL VALUE, and by first variable (S210) that can obtain those proper vectors behind the step S206, in step S208, should relevant group be after calculating, but be not limited to, one flow, one pressure, one temperature or a volume, and in step S210, this first variable is the pairing proper vector of eigenvalue of maximum, has maximal projection amount person on each input variable direction.
Except first variable that above-mentioned steps S210 is obtained, remaining system variable must be then via behind the step S206, a halt that determines those eigenwerts to calculate is a degree of freedom in system (S212), wherein this halt is for declining to a great extent some the second time of those eigenwerts, and this degree of freedom in system is the variable quantity of all eigenwerts before declining to a great extent some the second time of those eigenwerts.
In output terminal variable place, stipulate an estimating target rule (S214) of those output variables and those input variables, wherein the computing formula of this estimating target rule is as follows:
f(X)＝|a*CoC(X，Output)/b*CoC(X，Selected_Inputs)|....(1)
Or
f(X)＝||a*CoC(X，Output)|-|b*CoC(X，Selected_Inputs)||....(2)
In above-mentioned formula (1) or (2), this a and this b are adjustable variables, and this CoC is a correlation coefficient function, and this X is a variable to be selected, and Output is an output variable, the variable of Selected_Inputs for having selected.
The computing formula of above-mentioned estimating target rule only is embodiment, should not be interpreted as restriction of the present invention, as long as yet meet " system variable that is obtained and the system variable of having selected between relation heal little better (that is more independent better); and and relation is more better between the output variable " condition, the user is condition and define multiple estimating target rule according to this.
According to this estimating target rule, this first variable and this degree of freedom in system are that the basis is to obtain remaining variable till this variable quantity satisfies (S216), wherein the input of step S216 source comprises first variable (S210) step of above-mentioned those proper vectors of acquisition, carry out estimating target rule (S214) step that increases variable quantity (S224) step or stipulate those output variables and those input variables, the detailed process of step S216 please refer to Fig. 4, comprise and obtain this variable to be increased (S234), detecting this waits to increase variable and whether meets this estimating target rule (S236), if meet, then this waits to increase variable promptly becomes a suitable variable (S238), if do not meet, then continue to obtain this variable to be increased, and judge whether to reach this variable quantity (S240), if judged result is for being, then obtain the variable quantity (S218) of an initial proposed, otherwise, if judged result is then got back to the step that obtains this variable to be increased (S234) for not.
Obtain the variable quantity (S218) of an initial proposed, wherein the input of the variable quantity of this initial proposed source can be those input variables of this relevant group, then set up a system model (S220) of the variable quantity of this initial proposed, detect whether the error that this system model produces is to accept (S222), wherein should detect the step of this system model, if testing result is for being, then finish the Variables Selection flow process of this model construction, otherwise, if testing result more comprises and carries out the step that increases variable quantity (S224) for not.
About increasing the step of variable quantity (S224), please refer to shown in Figure 3, the input of this increase variable quantity source comprise decision systems degree of freedom (S212) and finish by the S222 step of above-mentioned Fig. 2 after enter S334 the makers-up of step institute, when the decision systems degree of freedom or after increasing the system variable number, then calculate explanation degree (S228), the condition (S230) that the comparison user sets, wherein this pre-conditioned be obtained person in the qualifications storehouse that sets of a user, in the step of this comparison, if comparison result is for being, then export this variable to be increased (S232) after comparison meets, as do not meet the minimum requirement of required explanation degree, promptly increase variable number recommended value (S334).
Whether an error that is produced in above-mentioned this system model of detection is can accept in (S222) step, when testing result is not, then entering increases system variable (S334) and carries out the step that increases variable quantity (S224), and the method flow of its increase variable quantity as shown in Figure 3.
Please refer to Fig. 5, appreciation of system model dimension and significant variable method for screening detail flowchart for second embodiment of the invention, comprise by the database that links on the line, reading system data to be processed (S300) in one database or any type of file, the user defines the required maximum variable quantity (S302) of this model construction, again these data are utilized the defined correlated condition of user those data that read to be carried out the action (S304) of variable classification, above-mentioned variable includes but not limited to flow, pressure or temperature etc., moreover, variable behind these variable classifications is divided into two kinds of input variable (S306) and output variables (S308), then, input variable is analyzed (S312) with STUDY ON PRINCIPAL VALUE (PCA) method, and the variable after will analyzing carries out the classification (S310) of different groups according to interdependence with input variable, and this different group for example classifies: temperature, volume, pressure or flow etc.Because of carrying out the action of STUDY ON PRINCIPAL VALUE, so input variable will produce at least one proper vector (S314) and at least one eigenwert (S316) after via STUDY ON PRINCIPAL VALUE, system is by finding out halt (S318) in those eigenwerts, the arrangement of those eigenwerts of being finished by above-mentioned STUDY ON PRINCIPAL VALUE is to be discharged to minimum value by maximal value, known PCA technology be according to the eigenwert that obtained explain number percent whether satisfy the user fixed condition, yet method of the present invention then can directly decline to a great extent a little as halt for the second time from these eigenwerts, decline to a great extent for the second time a little is from these eigenwerts proportionate relationship person of trying to achieve each other, and the account form that method of the present invention declines to a great extent a little the mentioned second time illustrates at this measure one example, if utilize the STUDY ON PRINCIPAL VALUE method to try to achieve out y
nIndividual eigenwert, wherein n is 1 to 10 value, eigenwert is the arrangement mode that successively decreases, and the account form that declines to a great extent for the second time a little is to come out with percentage calculation, as first group of y
1-y
2/ y
1Resulting number percent is 5%, and second group of y
2-y
3/ y
2Resulting number percent is 0%, but the 3rd group of y
3-y
4/ y
3Resulting number percent is 20%, and the 3rd group that so can get declines to a great extent a little for the present invention the so-called second time.
Then, by before the halt of above-mentioned S318 step the eigenwert that obtains with decision systems variable quantity initial value (S320), again the proper vector that these variable quantity initial values and step 314 are obtained converge whole together, for instance, if the user provides required variable total amount in step S302 be 20, the variable classification that carries out by above-mentioned steps S304, those proper vectors of those input variables (S306) and step S312 compare (S322), via obtaining maximum variable after the comparison, this variable is first variable (S324), the system variable quantity that system will be determined from step S320 is singly carried out variable in regular turn and is explained PCA degree of variation (S326), explanation degree in step S326 is by the definien of institute of system, set up qualifications storehouse (S328) this user in prior to system, above-mentioned qualifications storehouse comprises the explanation degree value of the selected variable of definition to total system.
The variable of qualifications and step S326 is done a comparison action (S330), if comparison result is when meeting, just selected variable is become suggestion choice variable number (S332), suppose that step (S332) obtains to advise that variable quantity is 4 thus, because obtained first variable by step S324, therefore the system variable that the demand of going back at present gets is 3, otherwise, when if comparison result does not meet, system will increase system variable number (S334) again from step S320, system is in order to try to achieve remaining variable, system will define an estimating target rule to select the remaining variable (S336) that meets the estimating target rule in regular turn after step S328, above-mentioned estimating target rule is that selected system variable and output variable concern maximum, and concern reckling with input variable, the formula of its calculating is exposed in, no longer repeats at this.
According to the estimating target rule from the only variable of choosing of step S332 (S338), judge then whether the system variable number satisfies (S340), if judged result is for being, then can obtain system's initial proposed variate-value (S342), wherein the input of the variable quantity of this initial proposed source can be should be correlated with those input variables of group of step S310, so system can use by finding out close variable in the relevant group, otherwise, if the judged result of S340 step is for denying, then get back to the step of estimating target value (S348), calculate after the desired value, get back to step S338, system will set up system model (S344) according to system's initial proposed variate-value that step S342 is obtained, and the system model of being set up can be neural network, FUZZY NETWORK or the pattern of setting up according to alternate manner.Accuracy in order to ensure system model, method of the present invention will be to the action (S346) of the system model operation mode error set up, when the system model of setting up when system discovery produces error, can judge whether the system model error is system's tolerance interval, if error for can accept the time, just finishes whole flow process, otherwise, if when error was unacceptable, step S334 will get back to further increase system variable number in system.
Please refer to Fig. 6, system schematic for system model dimension of the present invention appreciation and significant variable screening, system 1 of the present invention can be applicable to a computer, and the electric equipment 2 that the outside links is provided the variable function of a model construction, this system 1 comprises an input data module 10, it can obtain a plurality of data by a Data Source place, and this Data Source place is by the database that links on the line, a database or any type of file.One variable quantity module 12 is electrically connected at this input data module 10, defines the required maximum variable quantity of this model construction in order to a user.One STUDY ON PRINCIPAL VALUE module 16, it can carry out those input variables STUDY ON PRINCIPAL VALUE to obtain at least one proper vector and at least one eigenwert.One variable classification module 14 is electrically connected at this variable quantity module 12, and it can be categorized as at least one input variable and at least one output variable according to the defined correlated condition of this user with those data.
One relevant group module 18 will be carried out group's classification via those input variables behind this STUDY ON PRINCIPAL VALUE module analysis.One first variable module 20, its can by eigenvalue of maximum in those eigenwerts the maximum proper vector of corresponding those proper vectors.One degree of freedom in system module 22, it can be by declining to a great extent the second time in those eigenwerts o'clock as a degree of freedom in system and the variable quantity of all eigenwerts before obtaining second time in those eigenwerts to decline to a great extent a little.
One variable number computing module 24, be electrically connected at this degree of freedom in system module 22, it can be caught by this degree of freedom in system module 22 and get at least one variable to be increased, this variable number computing module 24 comprises a qualifications storehouse 240, it stores a plurality of qualificationss in advance, in order to limit variable to be selected, one explains degree unit 242, its choose this degree of freedom in system module those variable quantity a variable and explain the STUDY ON PRINCIPAL VALUE degree of variation of this selected variable, one comparing unit 244, it makes comparisons those qualificationss and selected this variable, and an output variable unit 246, and it will meet the variable of this comparing unit after relatively and export.
One choice variable module 26, be electrically connected at this variable number computing module 24, it can be by this variable number computing module 24, a plurality of choice variables that this first a variable module 20 and an estimating target unit 260 are obtained for the basis, this choice variable module 26 comprises an estimating target unit 260, it chooses an estimating target value of variable relation minimum in order to definition with those output variable relation maximums and with those, and a choice variable unit 262, it is according to this estimating target value, this first variable module and this degree of freedom in system module are by selecting at least one variable in this variable number computing module.
One judgment variable module 28, be electrically connected at this relevant group module 18, whether it equals this variable quantity module 14 default these variable quantity in order to the quantity of judging those choice variables that this choice variable module 26 is obtained, one sets up mode module 30, its those choice variables that this choice variable module can be obtained are set up pattern, one mode error module 32, it can be in order to adjust this degree of freedom in system, then, promptly can be linked to this variable number computing module 24 to increase required system variable if this mode error module produces when error is arranged.
By the above embodiments as can be known, method of the present invention can adopt the input of various data informations, can reach automatic analysis decision systems degree of freedom and the result who screens significant variable by above-mentioned method flow of the present invention.System of the present invention can be applicable to a computer, and the electric equipment that the outside links is provided the variable function of a model construction, can be accurately according to the system model that is obtained operating this electric equipment, method of the present invention can be applicable to various electric equipments and comprises semiconductor equipment or household appliances equipment to promote system's accuracy.
The present invention really can borrow above-mentioned disclosed technology, provides a kind of far different in known design, may be able to improve whole use value, does not see publication or public use before its application again, meets the important document of patent of invention, proposes application for a patent for invention in accordance with the law.
But, above-mentioned disclosed accompanying drawing, explanation, only be embodiments of the invention, all one of ordinary skilled in the art does other all equivalence improvement that belongs to invention spirit of the present invention according to above-mentioned explanation, the claim that still belongs to the present invention and defined.
Claims (26)
1. system model dimension appreciation and significant variable method for screening is characterized in that, comprising:
Read a plurality of data;
Utilize the defined correlated condition of a user that the data that read are carried out a variable classification, this variable classification comprises a plurality of output variables and a plurality of input variable that this user will preset;
Utilize at least one proper vector and at least one eigenwert of STUDY ON PRINCIPAL VALUE method in the hope of those input variables;
A halt that determines those eigenwerts to calculate is a degree of freedom in system;
Obtain first variable of those proper vectors;
Stipulate an estimating target rule of those output variables and those input variables;
According to this estimating target rule, this first variable and this degree of freedom in system is that the basis is to obtain remaining variable till required variable quantity satisfies;
Obtain the variable quantity of an initial proposed; And
Set up a system model of the variable quantity of this initial proposed.
2. appreciation of the system as claimed in claim 1 pattern dimension and significant variable method for screening is characterized in that, those variable classifications are a flow, a pressure, a temperature, a volume.
3. appreciation of the system as claimed in claim 1 pattern dimension and significant variable method for screening, it is characterized in that, more comprise and set a relevant group, to carry out group's classification via those input variables that this STUDY ON PRINCIPAL VALUE method is tried to achieve, wherein should relevant group be a flow, a pressure, a temperature, a volume after calculating.
4. appreciation of the system as claimed in claim 1 pattern dimension and significant variable method for screening is characterized in that, wherein those data are obtained by a database.
5. appreciation of the system as claimed in claim 1 pattern dimension and significant variable method for screening is characterized in that, wherein carry out the step of a variable classification, comprise that more this user defines the required maximum variable quantity of this system model construction.
6. appreciation of the system as claimed in claim 1 pattern dimension and significant variable method for screening is characterized in that, wherein this halt is for declining to a great extent some the second time of those eigenwerts.
7. appreciation of the system as claimed in claim 1 pattern dimension and significant variable method for screening is characterized in that, wherein this degree of freedom in system is the variable quantity of all eigenwerts before declining to a great extent some the second time of those eigenwerts; This first variable is the pairing proper vector of eigenvalue of maximum.
8. appreciation of the system as claimed in claim 1 pattern dimension and significant variable method for screening is characterized in that, wherein this estimating target rule is tried to achieve by following formula:
F (X)=| a*CoC (X, Output)/b*CoC (X, Selected_Inputs) | or
f(X)＝||a*CoC(X，Output)|-|b*CoC(X，Selected_Inputs)||
Wherein this a and this b are adjustable variables, and this CoC is a correlation coefficient function, and this X is a variable to be selected.
9. system model dimension as claimed in claim 8 appreciation and significant variable method for screening, it is characterized in that, wherein this estimating target rule for meet relation between the system variable that is obtained and the variable of having selected heal better little, and and the estimating target rule that condition defined more better of the relation between the output variable.
10. system model dimension as claimed in claim 3 appreciation and significant variable method for screening is characterized in that, wherein the input of the variable quantity of this initial proposed source is those input variables of this relevant group.
11. dimension appreciation of the system as claimed in claim 1 pattern and significant variable method for screening is characterized in that, wherein this sets up the variable quantity step of this initial proposed, more comprises detecting whether the error that this system model produced is to accept.
12. system model dimension as claimed in claim 11 appreciation and significant variable method for screening, it is characterized in that, wherein should detect the step of this system model, if testing result is for being, then finish the Variables Selection flow process of this system model construction, otherwise, if testing result more comprises and carries out the step that increases variable quantity for not.
13. dimension appreciation of the system as claimed in claim 1 pattern and significant variable method for screening is characterized in that, wherein this system model is neural network or FUZZY NETWORK.
14. the method for the increase variable quantity of a system model dimension as claimed in claim 12 appreciation and significant variable method for screening is characterized in that this method comprises:
Catch and get one of this degree of freedom in system and wait to increase variable or increase by a system variable number;
Explain that this waits to increase a STUDY ON PRINCIPAL VALUE degree of variation of variable;
Compare pre-conditioned whether the meeting that this waits to increase variable and this system; And
Output meets this this pre-conditioned variable to be increased after comparison.
15. the method for increase variable quantity as claimed in claim 14 is characterized in that, wherein obtained person in this pre-conditioned qualifications storehouse that is set by a user.
16. the method for increase variable quantity as claimed in claim 14, it is characterized in that, wherein this comparison this wait to increase in the step of variable, if comparison result is for being, then export after comparison and meet this this pre-conditioned variable to be increased, otherwise, if comparison result for not, then continues to catch and gets one of this degree of freedom in system and wait to increase variable or increase by a system variable number.
17. the method for remaining variable of acquisition of the system as claimed in claim 1 pattern dimension appreciation and significant variable method for screening is characterized in that this method comprises:
Obtain a variable to be increased;
Detecting this waits to increase variable and whether meets this estimating target rule;
If meet, then this waits to increase variable promptly becomes a suitable variable;
If do not meet, then continue to obtain this variable to be increased; And
Judge whether to reach this variable quantity.
18. the method for remaining variable of acquisition as claimed in claim 17 is characterized in that, wherein the step that should judge, if judged result for being, then obtains the variable quantity of this initial proposed, otherwise, if for not, then getting back to, judged result obtains a step of waiting to increase variable.
19. the system of system model dimension appreciation and significant variable screening, it is applied to a computer, and the variable function of a system model construction can be provided the electric equipment that the outside links, and it is characterized in that this system comprises:
One input data module, it can obtain a plurality of data by a Data Source place;
One variable quantity module is electrically connected at this input data module, uses for a user and defines the required maximum variable quantity of this system model construction;
One variable classification module is electrically connected at this variable quantity module, and it can be categorized as at least one input variable and at least one output variable according to the defined correlated condition of this user with those data;
One STUDY ON PRINCIPAL VALUE module, it can carry out those input variables STUDY ON PRINCIPAL VALUE to obtain at least one proper vector and at least one eigenwert;
One degree of freedom in system module, it can be by declining to a great extent the second time in those eigenwerts o'clock as a degree of freedom in system and the variable quantity of all eigenwerts before obtaining second time in those eigenwerts to decline to a great extent a little;
One variable number computing module is electrically connected at this degree of freedom in system module, and it can be caught by this degree of freedom in system module and get at least one variable to be increased;
One choice variable module is electrically connected at this variable number computing module, and it can be that the basis obtains a plurality of choice variables by this variable number computing module, one first variable module and an estimating target unit; And
One sets up mode module, and its those choice variables that this choice variable module can be obtained are set up this system model.
20. the system of system model dimension as claimed in claim 19 appreciation and significant variable screening is characterized in that wherein this Data Source place is a database.
21. the system of system model dimension as claimed in claim 19 appreciation and significant variable screening is characterized in that, this first variable module can obtain corresponding maximum proper vector by the eigenvalue of maximum in those eigenwerts.
22. the system of system model dimension as claimed in claim 19 appreciation and significant variable screening is characterized in that wherein this variable number computing module more comprises:
One qualifications storehouse, it stores a plurality of qualificationss in advance, in order to limit variable to be selected;
One explains the degree unit, and it is chosen in those variable quantity of this degree of freedom in system module this and waits to increase variable and explain that selected this wait to increase the STUDY ON PRINCIPAL VALUE degree of variation of variable;
One comparing unit, it makes comparisons those qualificationss and selected this variable to be increased; And
One output variable unit, it exports this variable to be increased that this comparing unit meets those qualificationss after relatively.
23. the system of system model dimension as claimed in claim 19 appreciation and significant variable screening is characterized in that wherein this choice variable module more comprises:
One estimating target unit, its in order to definition with those output variables relation maximum and with those estimating target rules of choosing the variable relation minimum to obtain an estimating target value; And
One choice variable unit, its according to this estimating target value, this first variable module and this degree of freedom in system module by selecting at least one choice variable in this variable number computing module.
24. the system of system model dimension as claimed in claim 19 appreciation and significant variable screening is characterized in that, more comprises a relevant group module, will carry out group's classification via those input variables behind this STUDY ON PRINCIPAL VALUE module analysis.
25. the system of system model dimension as claimed in claim 24 appreciation and significant variable screening, it is characterized in that, more comprise a judgment variable module, be electrically connected at this relevant group module, whether it equals default this variable quantity of this variable quantity module in order to the quantity of judging those choice variables that this choice variable module is obtained.
26. the system of system model dimension as claimed in claim 19 appreciation and significant variable screening is characterized in that more comprise a mode error module, it can be in order to adjust this degree of freedom in system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200610154140XA CN101145030B (en) | 2006-09-13 | 2006-09-13 | Method and system for increasing variable amount, obtaining rest variable, dimensionality appreciation and variable screening |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200610154140XA CN101145030B (en) | 2006-09-13 | 2006-09-13 | Method and system for increasing variable amount, obtaining rest variable, dimensionality appreciation and variable screening |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101145030A CN101145030A (en) | 2008-03-19 |
| CN101145030B true CN101145030B (en) | 2011-01-12 |
Family
ID=39207591
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200610154140XA Expired - Fee Related CN101145030B (en) | 2006-09-13 | 2006-09-13 | Method and system for increasing variable amount, obtaining rest variable, dimensionality appreciation and variable screening |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101145030B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103942403B (en) * | 2013-01-22 | 2017-07-14 | 阿里巴巴集团控股有限公司 | A kind of method and apparatus screened to magnanimity variable |
| CN103413052B (en) * | 2013-08-21 | 2016-02-24 | 华北电力大学 | Based on thermal power plant's Steady-state process modeling Variable Selection method of correlation analysis |
| US9891913B2 (en) * | 2014-12-23 | 2018-02-13 | Intel Corporation | Method and apparatus for performing conflict detection using vector comparison operations |
| CN108090241B (en) * | 2016-11-23 | 2021-08-17 | 财团法人工业技术研究院 | Trend variable identification method and system of continuous process |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2003060652A2 (en) * | 2002-01-15 | 2003-07-24 | Vanderbilt University | Method and apparatus for multifactor dimensionality reduction |
| CN1578955A (en) * | 2001-09-04 | 2005-02-09 | 国际商业机器公司 | Sampling approach for data mining of association rules |
| CN1794621A (en) * | 2006-01-12 | 2006-06-28 | 北京大学 | Construction method of non-regular permutation matrix LDPC code and its device |
| CN1808949A (en) * | 2005-12-23 | 2006-07-26 | 西安交通大学 | Non-physical modeling and emulation method for channels in multi-input and multi-output communication system |
-
2006
- 2006-09-13 CN CN200610154140XA patent/CN101145030B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1578955A (en) * | 2001-09-04 | 2005-02-09 | 国际商业机器公司 | Sampling approach for data mining of association rules |
| WO2003060652A2 (en) * | 2002-01-15 | 2003-07-24 | Vanderbilt University | Method and apparatus for multifactor dimensionality reduction |
| CN1808949A (en) * | 2005-12-23 | 2006-07-26 | 西安交通大学 | Non-physical modeling and emulation method for channels in multi-input and multi-output communication system |
| CN1794621A (en) * | 2006-01-12 | 2006-06-28 | 北京大学 | Construction method of non-regular permutation matrix LDPC code and its device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101145030A (en) | 2008-03-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chien et al. | Manufacturing intelligence to forecast and reduce semiconductor cycle time | |
| US11029359B2 (en) | Failure detection and classsification using sensor data and/or measurement data | |
| CN101458515B (en) | Wafer quality analysis method | |
| TWI549007B (en) | Method for searching and analyzing process parameters and computer program product thereof | |
| US7363195B2 (en) | Methods of configuring a sensor network | |
| CN114298863B (en) | Data acquisition method and system of intelligent meter reading terminal | |
| CN112115306B (en) | Method and system for performing automatic root cause analysis of anomalous events in high dimensional sensor data | |
| Sibalija et al. | An integrated approach to optimise parameter design of multi-response processes based on Taguchi method and artificial intelligence | |
| Pan et al. | A virtual metrology system for predicting end-of-line electrical properties using a MANCOVA model with tools clustering | |
| US11022642B2 (en) | Semiconductor yield prediction | |
| Xu et al. | Hybrid feature selection for wafer acceptance test parameters in semiconductor manufacturing | |
| CN111291783A (en) | Intelligent fault diagnosis method, system, terminal and storage medium for gas pressure regulating equipment | |
| CN101145030B (en) | Method and system for increasing variable amount, obtaining rest variable, dimensionality appreciation and variable screening | |
| Cherkassky et al. | Prediction of electric power consumption for commercial buildings | |
| KR20140113153A (en) | Method and System for Statistical Equivalence Test | |
| Susto et al. | A virtual metrology system based on least angle regression and statistical clustering | |
| Sim | Big data analysis methodology for smart manufacturing systems | |
| KR20010033059A (en) | Process for monitoring processing plants | |
| Li et al. | Using intelligent technology and real-time feedback algorithm to improve manufacturing process in IoT semiconductor industry | |
| KR102574244B1 (en) | A method and apparatus for generating future demand forecast data based on attention mechanism | |
| Wang et al. | A multivariate sign chart for monitoring dependence among mixed-type data | |
| CN116227896A (en) | Silicon carbide production process management method and system | |
| US20080125883A1 (en) | Method and apparatus for consolidating process control | |
| CN113610575B (en) | Product sales prediction method and prediction system | |
| TWI294971B (en) | Method of system model dimension identification and important variables selection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110112 Termination date: 20150913 |
|
| EXPY | Termination of patent right or utility model |