CN102254184B - Method for fusing multi-physical-domain feature information - Google Patents
Method for fusing multi-physical-domain feature information Download PDFInfo
- Publication number
- CN102254184B CN102254184B CN 201110200365 CN201110200365A CN102254184B CN 102254184 B CN102254184 B CN 102254184B CN 201110200365 CN201110200365 CN 201110200365 CN 201110200365 A CN201110200365 A CN 201110200365A CN 102254184 B CN102254184 B CN 102254184B
- Authority
- CN
- China
- Prior art keywords
- probability
- characteristic information
- interval
- generalized
- engineering
- 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
Images
Abstract
The invention discloses a method for fusing multi-physical-domain feature information, and the method comprises the following steps of: (1) obtaining required apriori information in an engineering event through a measuring tool, removing redundant information and obtaining an optimized feature information set; (2) converting each feature information value Xi in the optimized feature information into an interval form; (3) solving a broad sense Hidden Markov initial model; (4) obtaining an optimal model according to the obtained broad sense Hidden Markov initial model alpha; and (5) solving the optimum broad sense Hidden Markov model of each single-physical-domain, after synthesizing combined broad sense interval, checking the propability distribution according to a coupling relationship among various single-physical-domain optimum model parameters, and finishing and obtaining the fusion of the multi-physical-domain feature information. The method can be used for establishing coupling relationship among the multi-physical-domain parameters in different engineering and effectively solving the fusion problem of the multi-physical-domain feature information.
Description
Technical field
The present invention relates to field of information processing, particularly a kind of method of many physical domain feature fusion.
Background technology
Many physical domain feature fusion is to process most important components at engineering information, is to utilize the technology such as computing machine that all observation information are carried out automatic analysis, comprehensively finished decision-making and estimation task.Many physical domain feature fusion often can run into two class problems, and the one, the characteristic information uncertain problem that causes owing to the scarcity of the imperfection of the inexactness of survey instrument, information and randomness and priori; The 2nd, how many physical domain (such as vibration, temperature, noise and the cutting force etc.) characteristic information during engineering information is processed merges problem.
The characteristic information uncertainty is inevitable in Engineering Modeling and the many physical domain observation process, wants to find the characteristic parameter of one group of energy accurate description engineering state and performance, sets up identification and the forecast model of this stack features parameter, is a very difficult job.Although many researchists have spent the plenty of time and carried out constantly effort, effect is little.To the research method of uncertain problem, mostly described by the accurate probability of observational variable in the past, and only considered the randomness of observation, and ignored the imperfection of observation information and the scarcity of priori, this obviously can lower result's degree of belief.In recent years, people attempt going to address this problem with Imprecise probability, and begin to be applied to engineering field, such as Data Fusion of Sensor, reliability estimation, reliability optimal design, and uncertain design proposal etc., the lower bound of its out of true interval probability semantically can't form closure strictly less than the upper bound in interval, reasoning and demonstrating is complicated, calculates to be difficult to process.
Many physical domain feature fusion also is the typical problem during engineering information is processed, present research method more is for determining amount, research for random information and Incomplete information is less, wherein, relatively typical method mainly contains two classes: a class is the multiple dimensioned gaussian variable model that Choi M.J etc. proposes, Monte Carlo (Monte Carlo) simulation that Chamoin L etc. propose, method of weighted mean, Kalman filtering method, the multi-Bayes estimation technique and D-S evidential reasoning method, this several method all is the probability distribution of having supposed that first certain is special, be in the same place so that the scarcity of the randomness of observation and priori is obscured, thereby limited its application; An other class research is based on Hidden Markov Model (HMM) and related expanding model thereof, wherein, use Hidden Markov Model (HMM) and will solve 3 basic problems: evaluation problem, decoding problem and training problem, specific implementation adopts forward backward algorithm, Viterbi algorithm and Baum-Welch algorithm to finish, and sees " the A tutorial on Hidden Markov Models and selected applications in speech recognition " of Lawre R.Rabiner for details.Hidden Markov Model (HMM) is described information implicit in the engineering by the observation that is easy to the observation vector sequence, and the coupled relation between implicit information is processed many physical domain information fusion problem in the observation vector sequence by different physical quantities and the engineering, calculating is difficult to process, and Hidden Markov Model (HMM) can not solve the uncertain problem that the scarcity of the imperfection of information and priori is brought.
In the research in the past, what how characteristic information uncertainty and characteristic information in many physical domain feature fusion were merged usually all is independently, can not form an organic combination, causes the reliability, intelligent low of information processing.One or a class problem can only be solved, the problem that information uncertainty and many physical domain feature fusion combines can not be solved simultaneously.
Summary of the invention
The objective of the invention is to process for existing engineering information the deficiency of research method, a kind of many physical domain feature fusion method is provided, can solve that the characteristic information uncertain problem can solve many physical domain feature fusion problem again simultaneously in the engineering.
Realize that the concrete technical scheme that purpose of the present invention adopts is as follows:
A kind of many physical domain feature fusion method specifically comprises the steps:
(1) obtains the characteristic information collection
Obtain needed prior imformation in the engineering by survey instrument, remove redundant information, obtain the characteristic information collection of optimizing;
(2) each the characteristic information value X that step (1) characteristic information is concentrated
kConvert interval form to
Consider in the survey instrument measuring process and other uncertainty, by theory of errors each characteristic information value X
kConvert interval form to, to increase the reliability of the characteristic information value of measuring, wherein k is the sequence number of arbitrary characteristic information.
(3) ask for Generalized Implicit Markov initial model
At first according to the engineering actual conditions, divide engineering state;
Then utilize the interval characteristic information value X that has changed
kAnd the engineering state of dividing, with asking for Hidden Markov initial model similar approach, ask for state transition probability matrix A in the Generalized Implicit Markov initial model, observation probability matrix B and ask for original state probability matrix π according to check, wherein, probability replaces with the generalized interval probability in above-mentioned all matrixes, can obtain Generalized Implicit Markov initial model λ=(A, B, π);
Wherein, General Hidden Markov Model is with interval replacing characteristic information value in the Hidden Markov Model (HMM), replace probability in the Hidden Markov Model (HMM) with the generalized interval probability, and the model that organically combines with Hidden Markov.
The theoretical foundation of generalized interval probability is the Kaucher algorithm in the generalized interval.Upper and lower dividing value size in its interval probability is not limited greater than floor value by dividing value, and upper dividing value is less than or equal to floor value and all permits, semantic sealing;
The generalized interval probability satisfies the logical consistency constraint, has multiple possibility E such as an event
i, upper dividing value addition and the floor value addition result of all possible interval probability all are necessary for 1, namely
Keep logic consistent with the accurate probability of classics.
(4) according to Generalized Implicit Markov initial model λ and algorithm, obtain optimization model.
(4.1) utilize Generalized Implicit Markov initial model, and the concentrated observation sequence O of characteristic information, by the forward-backward algorithm algorithm, calculate the interval probability P (O| λ) under model λ.
(4.2) utilize Generalized Implicit Markov initial model, and the concentrated observation sequence O of characteristic information, by the viterbi algorithm, the corresponding optimum state sequence of preference pattern λ Q=q
1q
2Q
T
(4.3) utilize Generalized Implicit Markov initial model, and the concentrated observation sequence O of characteristic information, by the Baum-Welch algorithm, progressively improve the initial model parameter, until till the upper and lower boundary of P (O| λ) interval probability all restrains, can obtain optimum General Hidden Markov Model parameter
(5) utilize the dual random procedure structure of Hidden Markov Model (HMM), and the generalized interval bayes rule is set up the coupled relation between each single physical domain optimization model parameter, the generalized interval posterior probability of synthetic associating distributes, and can obtain the result of many physical domain feature fusion.
Wherein, the generalized interval bayes rule has replaced probability in the classical bayes rule with the generalized interval probability, and it is defined as follows:
In the formula: A is and event E
iThe physical event of coupled relation is arranged, has in addition:
dualp(A|E
j)＝1-p(A
c|E
j)，p(A|E
j)+p(A
c|E
j)＝1；
Distribute according to associating generalized interval posterior probability obtained above, obtain the result of many physical domain feature fusion, can carry out objective and accurate evaluation to state and the performance of engineering, combine with the generalized transition probability matrix based on the trend of historical data, state and the performance of engineering indicated.
The present invention compared with prior art, how information uncertainty problem and many physical domain characteristic information merge theory, method and the technology that provides new in the feature fusion in order to solve simultaneously, and following advantage is arranged:
1) the generalized interval probabilistic method had both been considered the randomness in the uncertain event, had considered again the imperfection of observation information and the scarcity of priori.
2) the generalized interval probabilistic method provides a kind of effective calculation method to be used for the uncertainty of engineering quantitative information and knowledge, the interval border that need not in the classic method is estimated, and support probability inference, in the situation of observation data deficiency, priori shortage, parameter distribution the unknown, also can obtain preferably result.
3) the generalized interval bayes rule combines with Hidden Markov Model (HMM), both can capture dependence and coupled relation between uncertain factor, has again very strong extensibility and versatility.By the generalized interval bayes rule, set up the coupled relation between many physical domain parameter in the different engineerings, can effectively solve many physical domain character information fusion problem.
Description of drawings
Accompanying drawing 1 is the FB(flow block) of many physical domain of the present invention feature fusion method embodiment.
Accompanying drawing 2 is FB(flow block) of embodiment of the invention machine tooling process status monitoring
Embodiment
The present invention will be by being further described take the monitoring of machine tooling process status as example embodiment.
Referring to accompanying drawing 2,
1) obtains many physical domain characteristic information of machine tooling
By sensor machine tooling process dynamics signal is tested, obtain the observational characteristic information (vibration, noise, temperature, cutting force, cutting speed, cutting depth etc.) of machine tooling, remove redundant information by methods such as neural network or wavelet analysises, make up the feature set that reflection machine tooling state or dynamic property are optimized;
Present embodiment selects sensor as the survey instrument of measuring characteristic information owing to be used for the machine tooling process.
2) with each characteristic information value X of machine tooling characteristic information value
kConvert intervalization form to
Consider the uncertainty in sensor error and the measuring process, by the error blending theory in the theory of errors or 3 σ rules the machine tooling characteristic information value of having optimized is converted to interval form, increase reliability.Machine tooling characteristic information value X
kCarry out intervalization and can adopt following method to realize, wherein k is the sequence number of arbitrary machine tooling characteristic information:
At first, according to becoming known for measuring characteristic information value X
kThe error of sensor be ε
k, try to achieve resultant error according to the error blending theory and be
Wherein: a
kBe the propagation of error coefficient, lathe machining feature value of information X is then arranged
kThe interval be [X
k-σ, X
kσ].
In addition, if the characteristic information value that the machine tooling characteristic information is concentrated is normal distribution, can also ask first standard deviation, then according to asking for 3 σ rules, the interval [X of being of lathe machining feature information data be arranged then
k-3 σ, X
k3 σ].
3) ask for the Generalized Implicit Markov initial model of machine tooling
At first according to the machine tooling actual conditions, divide the machine tooling state.Then utilize the interval machine tooling characteristic information value X that has changed
kWith asking for Hidden Markov initial model similar approach, ask for state transition probability matrix A in the Generalized Implicit Markov initial model, observation probability matrix B and ask for original state probability matrix π according to check, wherein, probability replaces with the generalized interval probability in above-mentioned all matrixes, can obtain Generalized Implicit Markov initial model λ=(A, B, the π) of machine tooling;
Wherein, General Hidden Markov Model is with interval replacing characteristic information value in the Hidden Markov Model (HMM), replace probability in the Hidden Markov Model (HMM) with the generalized interval probability, and the model that organically combines with Hidden Markov.
The generalized interval probability is the popularization of generalized interval and interval probability, namely the interval upper and lower boundary in the generalized interval is replaced with the probability size, its theoretical foundation is the Kaucher algorithm in the generalized interval, the interval has been converted to Probability Forms, upper and lower dividing value size in its interval probability, not limited greater than floor value by dividing value, upper dividing value is less than or equal to floor value and all permits, semantic sealing;
The generalized interval probability satisfies the logical consistency constraint, has multiple possibility E such as the machine tooling state
i, upper dividing value addition and the floor value addition result of the interval probability of all possible states all are necessary for 1, namely
Keep logic consistent with the accurate probability of classics.
4) according to Generalized Implicit Markov initial model λ and algorithm, obtain the optimization model of machine tooling.
(1) utilize Generalized Implicit Markov initial model, and the concentrated observation sequence O of machine tooling characteristic information, by the forward-backward algorithm algorithm, calculate the interval probability P (O| λ) under model λ.
(2) utilize Generalized Implicit Markov initial model, and the concentrated observation sequence O of machine tooling characteristic information, by the viterbi algorithm, the corresponding optimum state sequence of preference pattern λ Q=q
1q
2Q
T
(3) utilize Generalized Implicit Markov initial model, and the concentrated observation sequence O of machine tooling characteristic information, by the Baum-Welch algorithm, progressively improve the initial model parameter, (till the upper and lower boundary of (O| λ) interval probability all restrains, can obtain optimum machine tooling General Hidden Markov Model parameter until P
5) utilize the dual random procedure structure of Hidden Markov Model (HMM), and the generalized interval bayes rule is set up the coupled relation between each single physical domain optimization model parameter of machine tooling, synthetic bastard machine tool processing generalized interval posterior probability distributes, and obtains the result of many physical domain of machine tooling feature fusion.
Wherein, the generalized interval bayes rule has replaced probability in the classical bayes rule with the generalized interval probability, is defined as follows:
In the formula: E
iCorresponding event (the i=1 of machine tooling process status, ..., n, i are any one occurrence sequence number), A is the physical event (vibration, noise, temperature, cutting force, cutting speed, cutting depth etc.) that not machine tooling process status has coupled relation, has in addition:
dualp(A|E
j)＝1-p(A
c|E
j)，p(A|E
j)+p(A
c|E
j)＝1。
In the formula: E
jThe corresponding event of machine tooling process status, j is any one occurrence sequence number,
E
jThe probability supplementary set, its probability and be 1, A
c| E
jA|E
jThe probability supplementary set, its probability and be that 1, dual is the sign of operation that makes semantic sealing.
Distribute according to associating generalized interval posterior probability obtained above, obtain the result of many physical domain feature fusion, can carry out objective to the state of machine tooling and accurate evaluation is finished monitoring.With combine based on the state transition probability matrix in the lathe General Hidden Markov Model of historical data trend, also can the state of machine tooling be indicated.
In step 4) in, according to optimization model
Utilize the above-mentioned the 2nd) step, and the observation sequence O that concentrates of the characteristic information of observation can obtain the optimum state of single physical domain.
Above-described embodiment only is a preferred scheme of method of the present invention, method of the present invention is not limited to and the monitoring that is used for the machine tooling process status, Other Engineering field such as Intelligent Measurement, robot, many physical domain feature fusion such as automatic target identification are all applicable etc.
Claims (6)
1. the method for physical domain feature fusion more than a kind specifically comprises the steps:
(1) obtains the characteristic information collection of each single physical domain
Obtain needed prior imformation in the engineering event by survey instrument, remove redundant information wherein, obtain the characteristic information collection of optimizing;
(2) each characteristic information value X that the characteristic information that step (1) is obtained is concentrated
kConvert interval form to, wherein k is the sequence number of arbitrary characteristic information value;
(3) ask for Generalized Implicit Markov initial model, detailed process is:
At first according to the engineering actual conditions, divide engineering state;
Then utilize the interval characteristic information value X that has changed
kAnd the engineering state of dividing, ask for state transition probability matrix A, observation probability matrix B and original state probability matrix π, can obtain Generalized Implicit Markov initial model λ=(A, B, π), wherein, the probability in above-mentioned each probability matrix replaces with the generalized interval probability;
(4) according to the Generalized Implicit Markov initial model λ that obtains, obtain optimization model, be specially:
(4.1) the observation sequence O that utilizes characteristic information to concentrate by the forward-backward algorithm algorithm, calculates the interval probability P (O| λ) under described Generalized Implicit Markov initial model λ;
(4.2) the observation sequence O that utilizes characteristic information to concentrate by the viterbi algorithm, selects optimum state sequence Q=q corresponding to described Generalized Implicit Markov initial model λ
1q
2Q
T
(4.3) utilize described observation sequence O, by the Baum-Welch algorithm, adjust the parameter of described initial model λ, until till the upper and lower boundary of P (O| λ) interval probability all restrains, namely obtain optimum General Hidden Markov Model
(5) utilize above-mentioned steps to try to achieve the General Hidden Markov Model of the optimum of each single physical domain, set up coupled relation between each single physical domain optimization model parameter according to the generalized interval bayes rule again, the generalized interval posterior probability of synthetic associating distributes, and namely finishes to obtain many physical domain feature fusion.
2. method according to claim 1 is characterized in that, in the described step (2) with characteristic information value X
kThe interval form that converts to is [X
k-ε, X
k+ ε], wherein,
Be resultant error, a
kBe propagation of error coefficient, ε
kFor being used for measuring characteristic information value X
kThe error of survey instrument, wherein k is the sequence number of arbitrary characteristic information value.
3. method according to claim 1 is characterized in that, characteristic information value X in the described step (2)
kThe interval form that converts to is [X
k-3 σ, X
k+ 3 σ], at this moment, the characteristic information value that characteristic information is concentrated is normal distribution, and σ is its standard deviation.
4. one of according to claim 1-3 described method is characterized in that, described generalized interval bayes rule has replaced probability in the classical bayes rule with the generalized interval probability, and it is defined as follows:
Wherein, E
iBe i event in the engineering, i=1 ..., n, i are the event sequence number, and n is positive integer, and A is and event E
iThe physical event that coupled relation is arranged,
The summation symbol, in addition:
dualp(A|E
j)＝1-p(A
c|E
j)，
p(A?|E
j)+p(A
c|E
j)＝1；
In the formula, E
jBe j event in the engineering, j is any one occurrence sequence number,
E
jThe probability supplementary set, its probability and be 1, A
c| E
jA|E
jThe probability supplementary set, its probability and be that 1, dual is the sign of operation that makes semantic sealing.
5. one of according to claim 1-3 described method is characterized in that, described engineering event is machine tooling, Intelligent Measurement or automatic target identification.
6. one of according to claim 1-3 described method is characterized in that described survey instrument is sensor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110200365 CN102254184B (en) | 2011-07-18 | 2011-07-18 | Method for fusing multi-physical-domain feature information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110200365 CN102254184B (en) | 2011-07-18 | 2011-07-18 | Method for fusing multi-physical-domain feature information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102254184A CN102254184A (en) | 2011-11-23 |
| CN102254184B true CN102254184B (en) | 2013-01-02 |
Family
ID=44981436
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201110200365 Active CN102254184B (en) | 2011-07-18 | 2011-07-18 | Method for fusing multi-physical-domain feature information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102254184B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103034170B (en) * | 2012-11-27 | 2014-10-29 | 华中科技大学 | Numerical control machine tool machining performance prediction method based on intervals |
| CN107844806B (en) * | 2017-11-20 | 2021-06-11 | 华东交通大学 | Raisin color sorting method based on modal interval |
| CN108009573B (en) * | 2017-11-24 | 2020-08-14 | 北京物灵智能科技有限公司 | Robot emotion model generation method, emotion model and interaction method |
| CN108694417A (en) * | 2018-08-16 | 2018-10-23 | 湖南工程学院 | A kind of sliding bearing-rotor system amount of unbalance recognition methods |
| CN112171376B (en) * | 2020-08-21 | 2022-05-10 | 杭州玖欣物联科技有限公司 | Machine tool workpiece real-time statistical method based on current signal segmentation |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101520651B (en) * | 2009-03-03 | 2011-06-29 | 华中科技大学 | Analysis method for reliability of numerical control equipment based on hidden Markov chain |
| CN101984454B (en) * | 2010-11-19 | 2012-11-07 | 杭州电子科技大学 | Multi-source multi-characteristic information fusion method based on data drive |
-
2011
- 2011-07-18 CN CN 201110200365 patent/CN102254184B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN102254184A (en) | 2011-11-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103105820B (en) | Machining cutter abrasion state identification method of numerical control machine tool | |
| CN102254184B (en) | Method for fusing multi-physical-domain feature information | |
| CN103743402B (en) | A kind of underwater intelligent self adaptation Approach of Terrain Matching of topographic information based amount | |
| CN106677763B (en) | Dynamic integrated modeling-based oil well working fluid level prediction method | |
| CN105981025A (en) | Actually-measured marine environment data assimilation method based on sequence recursive filtering three-dimensional variation | |
| JP2021509211A (en) | How to design high-rigidity blades based on random isogeometric analysis | |
| CN103268525B (en) | A kind of Hydrological Time Series simulating and predicting method based on WD-RBF | |
| CN102842043B (en) | Particle swarm classifying method based on automatic clustering | |
| CN103065042A (en) | Multiple target comprehensive decision evaluation method based on scene | |
| CN105389326A (en) | Image annotation method based on weak matching probability canonical correlation model | |
| CN103577676A (en) | Grey weighting method for sewage treatment process comprehensive evaluation | |
| CN111709244A (en) | Deep learning method for identifying causal relationship of contradictory dispute events | |
| CN104795063A (en) | Acoustic model building method based on nonlinear manifold structure of acoustic space | |
| Chen et al. | Combining fuzzy iteration model with dynamic programming to solve multiobjective multistage decision making problems | |
| CN115755768A (en) | Five-axis linkage precision numerical control machine tool and control method thereof | |
| Roux et al. | An efficient parallel global optimization strategy based on Kriging properties suitable for material parameters identification | |
| CN101893430B (en) | Processing method of abnormal measured values based on CNC gear measuring center | |
| Wu et al. | Research on data reconciliation based on generalized T distribution with historical data | |
| Sun et al. | In-situ prediction of machining errors of thin-walled parts: an engineering knowledge based sparse Bayesian learning approach | |
| CN105373805A (en) | A multi-sensor maneuvering target tracking method based on the principle of maximum entropy | |
| CN112200252A (en) | Joint dimension reduction method based on probability box global sensitivity analysis and active subspace | |
| CN108197809B (en) | Real-time scheduling method of weight sharing deep network based on dimension optimal conversion | |
| CN106202694A (en) | Combination Kriging model building method based on combination forecasting method | |
| CN103425890A (en) | Landscape water quality analysis algorithm | |
| CN105575113A (en) | Sensing method of traffic running states |
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 |