CN1359646A - Method for creating fuzzy-neural network expert system for evaluating sensing quality of cigarette - Google Patents
Method for creating fuzzy-neural network expert system for evaluating sensing quality of cigarette Download PDFInfo
- Publication number
- CN1359646A CN1359646A CN 00129435 CN00129435A CN1359646A CN 1359646 A CN1359646 A CN 1359646A CN 00129435 CN00129435 CN 00129435 CN 00129435 A CN00129435 A CN 00129435A CN 1359646 A CN1359646 A CN 1359646A
- Authority
- CN
- China
- Prior art keywords
- neutral net
- training
- expert
- neural network
- user
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A method for creating a fuzzy-neural network expert system for evaluating the sensing quality of cigarettes includes classifying the cigarettes by experts while providing typical data, training the pointed neura network with said typical data as sample, creating neural networks relative to different classes, configuring artificial neural network (ANN) library, and managing it with particular program.
Description
The present invention relates to a kind of foundation, management and use of artificial neural network expert system, more particularly, relate to set up the fuzzy-neural network method of expert system of cigarette (comprising single-tobacco-typed cigarette and finished cigarettes) sensory quality assessment.
Traditional single-tobacco-typed cigarette and finished cigarettes sensory quality assessment are to carry out with the method for smokeing panel test of artificial (industry specialists).The shortcoming of this method is: expert's workload of smokeing panel test is big; Smoked panel test expert person and subjective factors such as healthy and mental status thereof of evaluation result influence greatly, so have tangible randomness, uniformity is poor.
At present, traditional artificial intelligence expert system successfully is applied to many aspects, and the field of deriving in the symbolic logic with accurate implication is very successful.But, there are some practical problems can't or to be difficult to describe with symbol technology or deterministic mathematical model, this is the ample scope for abilities of method such as neutral net, fuzzy set exactly.FUZZY SET APPROACH TO ENVIRONMENTAL has the ability of expressing human sense organ and thinking concept obfuscation; Neutral net is good at concluding from imperfect, as to contain noise input, extraction information, and can learn from the real world sample, obtains the neutral net expertise; Fuzzy-neutral net then has both advantages concurrently.
The object of the present invention is to provide a kind of minimizing manually smoke panel test workload, improve to estimate objective conforming cigarette (comprising finished cigarettes, single-tobacco-typed cigarette) sensory quality assessment method, promptly set up the fuzzy-neural network method of expert system of cigarette sensory quality assessment.
The present invention utilizes artificial neural network (ANN:Artificial Neural Network), study and popularization ability, the authentic data that provides with industry specialists is as sample, train a series of fuzzy-neutral net (storehouse); Well-organized, constitute man-machine interaction's formula, calculate the expert opinion system that combines with the traditional calculations machine technology based on fuzzy-nerve.This method is by industry specialists cigarette to be divided into groups, and typical data is provided, and makes sample with this typical data, the neutral net of training appointment; Set up many different neutral net one by one, constitute neutral net storehouse (ANN storehouse), and they manage to this with certain procedure by many different group correspondences.
Because in certain limit and degree, can partly replace expert's work of smokeing panel test with neural network method, therefore improved efficient greatly.For example, in the neutral net storehouse that trains, promptly in the scope of grasping its characteristic rule, as close tobacco production area, similar cigarette type, can not do the work of manually smokeing panel test, directly make knowledge and extract or promote and estimate, reduce the workload of manually smokeing panel test by neural network expert system.And the training sample of neutral net can be selected the typical data of smokeing panel test for use, promptly adopts the process of smokeing panel test strict, standard, the selected participation expert of work that smokes panel test, thus the objectively smoking result data that gain public acceptance are as training sample.So, can obtain to estimate comparatively accurately the expertise of single-tobacco-typed cigarette and finished cigarettes in the neutral net after the training, thereby realize therefore also having improved evaluation quality than tradition more reliable, the objective conforming evaluation of method of smokeing panel test.
Describe the present invention in detail below in conjunction with accompanying drawing and specific embodiment thereof.
Fig. 1 system program flow chart of the present invention.
Fig. 2 system of the present invention use and management programme diagram.
Fig. 3 system architecture of the present invention concerns schematic diagram.
Fig. 4 the present invention divides and the membership function curve map four classes of fuzzy quantity.
Fig. 5 the present invention divides and the membership function curve map five classes of fuzzy quantity.
Fig. 6 the present invention divides and the membership function curve map three classes of fuzzy quantity.
Fig. 3 shows an essential characteristic of the present invention, the structural relation of fuzzy-neural network expert system promptly of the present invention.Expert and user can use native system; The expert can import sample data with it, forms training sample set, specify also neural network training model, test neural network model, finally sets up fuzzy-neutral net experts database (ANN storehouse); The user can use native system to import the relevant data of task to be evaluated, calls neural network model, thereby is blured-sensory quality assessment result that neutral net expert provides.
Idiographic flow such as Fig. 1 that native system forms, it forms step or method is,
1. also arrangement sample data is collected in the grouping of industry specialists decision single-tobacco-typed cigarette or finished cigarettes evaluation;
2. certain group known sample data of industry specialists being submitted to write in the specific file, and it is carried out normalization and Fuzzy processing.Normalization scope and algorithm are decided according to the input concrete condition, and allow user and expert to make an amendment in corresponding program; If the sample data deficiency should organize industry specialists to smoke panel test, obtain new sample data;
3. system will relatively and find whether to have " special sample " (being unique sample with certain output parameter), and the prompting user determines whether to adopt this sample;
4. according to the input number of sample data, system specifies corresponding network model, can network parameter and definition learning rate and inertial parameter be set by system's setting or by user oneself;
5. utilize the corresponding neutral net of sample data training; In this network training, system is according to the learning state of training process display network, and in " training error " error change between display network output valve and the training sample set-point;
6. after this network training is finished, the neutral net that trains is deposited in the neutral net storehouse (ANN storehouse) of system;
7. said process can carry out repeatedly, checks whether will set up new group, and whether task is finished, sets up one by one up to whole groups to finish.
The division of cigarette group is as follows:
The cigarette type of single-tobacco-typed cigarette is divided: fire-cured tobacco type, air-curing of tobacco leaves type, suncured tabacco type.
The cigarette type of finished cigarettes is divided: fire-cured tobacco type, mixed type, outer odor type, cigar type etc.
Afterwards again in all types of by the place of production of pure tobacco leaf, position and color grouping, finished cigarettes is by the brand grouping, and each is with neutral net correspondence with it.
The basic quality and the physical and chemical index (will import as network) of thin material tobacco leaf are: the GB grade of pure tobacco leaf, the content of total nicotine, total reducing sugar, reduced sugar, total nitrogen, protein, chlorine etc.;
The basic quality of finished cigarettes and physical and chemical index (will as the network input parameter) be: the content of total nicotine, total reducing sugar, reduced sugar, total nitrogen, protein, chlorine etc., cigarette weight (g/ props up), total nitrogen resistance to suction (Pa), CO (mg/ props up), puff number/, nicotine (mg/ props up), moisture (mg/ props up), tar (mg/ props up);
Sensory evaluating smoking's parameter of single-tobacco-typed cigarette, promptly the output parameter as fuzzy-neutral net of single-tobacco-typed cigarette sensory quality assessment is: fragrance matter, perfume quantity, the property sent out, assorted gas, excitant, pleasant impression, strength and 7 odor types thoroughly: Luzhou-flavor, dense in perfume (or spice), middle giving off a strong fragrance, middle odor type, middle delicate fragrance, fragrant in clear, delicate fragrance type (dense, dense partially in, in dense partially, in, in clear partially, clear partially in, clearly)
Sensory evaluating smoking's parameter of finished cigarettes, promptly (will as network output parameter according to) be: fragrance, harmony, assorted gas, excitant, pleasant impression and gloss;
The obfuscation of parameter relates to the division of parameter value grade among the present invention, can by 3,4 and 5 etc. three kinds divide.It is divided and membership function figure such as Fig. 4, Fig. 5, shown in Figure 6.The requirement of neutral net in according to the present invention, and satisfy practical required precision, the data normalization scope is:
Former input data: (0.001,1) former output data: (0.001,0.99) for example: pure tobacco leaf grade is represented with international code name, seven parameters of each international code name correspondence, wherein preceding 5 parameters are fuzzy variable, and carry out divisions such as 3,4,5 respectively, the point of its degree of membership=1 is set as follows:
1). maturity (5 etc.)
Complete ripeness: 1.0 maturations: 0.70 is still ripe: 0.50 undercure: 0.30 is false ripe: 0.01
2). blade construction (4 etc.)
Loose: 0.01 is still loose: 0.375 is close slightly: 0.625 is tight: 1.0
3). identity (3 etc.)
Thick or broad 0.01 is thin slightly: 0.50 is medium: 1.0
4). oil content (4 etc.)
Heavy wool: 1.0 have oil: 0.625 has oil slightly: 0.375 few oil: 0.01
5). colourity (5 etc.)
Dense: the last 1.0: in 0.70: a little less than in the of 0.50: 0.30 is light: 0.01
6). length: normalization between 20 to 50.
7). permanent disability: normalization between 0.05 to 0.35.
The processing method of all fours is used for input, the output parameter of finished cigarettes.
Fig. 2 is the use and management program or the step of native system:
(1), press system prompt, the user imports single-tobacco-typed cigarette to be evaluated or finished cigarettes group number;
(2), system accesses the neutral net that has trained accordingly from the ANN storehouse;
(3), under system prompt, the user imports known basic quality parameter and physical and chemical index data;
(4), the neutral net that accesses does ' neural calculating ' (Neuro-computing), and with sense organ matter
Amount evaluation result (result of calculation) is converted into the understandable data mode output of user.
(5), above-mentioned steps repeatedly, finish up to appraisal.
Many feature and advantage of the present invention have been showed in the explanation of front, and the novelty that has in application.In addition, the skilled person in the present technique field, especially those technical staff that not only had practical experience but also had certain neutral net knowledge are not difficult some flow processs of native system are made an amendment.Therefore, the present invention is not limited to structure and operation illustrated and that described, the modification that all are suitable and substitute and all should be considered as falling into scope of the present invention.
Claims (3)
1, a kind of fuzzy-neural network method of expert system of setting up the cigarette sensory quality assessment is
(1) also arrangement sample data is collected in the grouping of industry specialists decision single-tobacco-typed cigarette or finished cigarettes evaluation;
(2) certain group known sample data of industry specialists being submitted to write in the specific file, and it is carried out normalization and Fuzzy processing, and allow user and expert to make an amendment in corresponding program; If the sample data deficiency should organize industry specialists to smoke panel test, obtain new sample data;
(3) system will relatively and find whether to have " special sample ", and the prompting user determines whether to adopt this sample;
(4) according to the input number of sample data, system specifies corresponding network model, and can network parameter and definition learning rate and inertial parameter be set by system's setting or by user oneself;
(5) utilize the corresponding neutral net of sample data training; In this network training, system is according to the learning state of training process display network, and in " training error " error change between display network output valve and the training sample set-point;
(6) after this network training is finished, the neutral net that trains is deposited in the neutral net storehouse of system;
(7) said process can carry out repeatedly, checks whether will set up new group, and whether task is finished, sets up one by one up to whole groups to finish.
2, expert system according to claim 1 is characterized in that the use and management program is
(1) press system prompt, the user imports single-tobacco-typed cigarette to be evaluated or finished cigarettes group number,
(2) system accesses the neutral net that has trained accordingly from the ANN storehouse,
(3) under system prompt, the user imports known basic quality parameter and physical and chemical index data,
(4) neutral net that accesses is done ' neural calculating ', and the sensory quality assessment result is converted into the understandable data mode output of user,
(5) above-mentioned steps repeatedly finishes up to appraisal.
3, expert system according to claim 1 and 2, the structural relation that it is characterized in that system is that expert and user can use native system: the expert can import sample data with it, forms training sample set, specify also neural network training model, test neural network model, finally sets up fuzzy-neutral net experts database; The user can use native system to import the relevant data of task to be evaluated, calls neural network model, thereby is blured-sensory quality assessment result that neutral net expert provides.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 00129435 CN1115112C (en) | 2000-12-21 | 2000-12-21 | Method for creating fuzzy-neural network expert system for evaluating sensing quality of cigarette |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 00129435 CN1115112C (en) | 2000-12-21 | 2000-12-21 | Method for creating fuzzy-neural network expert system for evaluating sensing quality of cigarette |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1359646A true CN1359646A (en) | 2002-07-24 |
| CN1115112C CN1115112C (en) | 2003-07-23 |
Family
ID=4593505
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 00129435 Expired - Lifetime CN1115112C (en) | 2000-12-21 | 2000-12-21 | Method for creating fuzzy-neural network expert system for evaluating sensing quality of cigarette |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1115112C (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100421585C (en) * | 2003-02-25 | 2008-10-01 | 中国海洋大学 | Method for establishing mixed expert system of maintaining cigarette leaf group formulation |
| CN100421586C (en) * | 2003-02-25 | 2008-10-01 | 颐中烟草(集团)有限公司 | Method for establishing mixed expert system of designing cigarette leaf group formulation |
| CN100565572C (en) * | 2005-09-28 | 2009-12-02 | 山东中烟工业公司 | Tobacco leaf quality management system and method |
| CN101502337B (en) * | 2009-02-26 | 2012-05-02 | 孟科峰 | Method for controlling model building in leaf moisture-regaining process of tobacco shred production |
| CN109034388A (en) * | 2018-07-27 | 2018-12-18 | 湖北中烟工业有限责任公司 | A kind of prediction model of cigarette material and mainstream smoke constituents based on Genetic Algorithm Optimized Neural Network |
| CN109330018A (en) * | 2018-10-30 | 2019-02-15 | 浙江中烟工业有限责任公司 | A kind of setting method of cigarette weight control system of cigarette making machine aspirator tape starting position |
| CN109993413A (en) * | 2019-03-04 | 2019-07-09 | 中国地质大学（武汉） | A kind of flue cured tobacco quality benefit integrated evaluating method and system based on data-driven |
| CN111523823A (en) * | 2020-05-07 | 2020-08-11 | 云南中烟工业有限责任公司 | Method for determining multi-point cigarette production standard sample through membership function |
| CN114460194A (en) * | 2022-01-26 | 2022-05-10 | 云南中烟工业有限责任公司 | Method for analyzing internal quality of tobacco mainstream smoke |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100454290C (en) * | 2005-11-28 | 2009-01-21 | 颐中烟草(集团)有限公司 | Cigarette organoleptic quality qualitative index estimating method |
| CN100444153C (en) * | 2005-11-28 | 2008-12-17 | 颐中烟草(集团)有限公司 | Cigarette internal quality index extimating method based on regression function estimating SVM |
| WO2019085369A1 (en) * | 2017-10-31 | 2019-05-09 | 高大启 | Electronic nose instrument and sensory quality evaluation method for tobacco and tobacco product |
-
2000
- 2000-12-21 CN CN 00129435 patent/CN1115112C/en not_active Expired - Lifetime
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100421585C (en) * | 2003-02-25 | 2008-10-01 | 中国海洋大学 | Method for establishing mixed expert system of maintaining cigarette leaf group formulation |
| CN100421586C (en) * | 2003-02-25 | 2008-10-01 | 颐中烟草(集团)有限公司 | Method for establishing mixed expert system of designing cigarette leaf group formulation |
| CN100565572C (en) * | 2005-09-28 | 2009-12-02 | 山东中烟工业公司 | Tobacco leaf quality management system and method |
| CN101502337B (en) * | 2009-02-26 | 2012-05-02 | 孟科峰 | Method for controlling model building in leaf moisture-regaining process of tobacco shred production |
| CN109034388A (en) * | 2018-07-27 | 2018-12-18 | 湖北中烟工业有限责任公司 | A kind of prediction model of cigarette material and mainstream smoke constituents based on Genetic Algorithm Optimized Neural Network |
| CN109330018A (en) * | 2018-10-30 | 2019-02-15 | 浙江中烟工业有限责任公司 | A kind of setting method of cigarette weight control system of cigarette making machine aspirator tape starting position |
| CN109330018B (en) * | 2018-10-30 | 2021-05-18 | 浙江中烟工业有限责任公司 | Method for setting starting position of cut tobacco suction belt of cigarette weight control system of cigarette making machine |
| CN109993413A (en) * | 2019-03-04 | 2019-07-09 | 中国地质大学（武汉） | A kind of flue cured tobacco quality benefit integrated evaluating method and system based on data-driven |
| CN111523823A (en) * | 2020-05-07 | 2020-08-11 | 云南中烟工业有限责任公司 | Method for determining multi-point cigarette production standard sample through membership function |
| CN114460194A (en) * | 2022-01-26 | 2022-05-10 | 云南中烟工业有限责任公司 | Method for analyzing internal quality of tobacco mainstream smoke |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1115112C (en) | 2003-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1115112C (en) | Method for creating fuzzy-neural network expert system for evaluating sensing quality of cigarette | |
| Aars et al. | Party politicisation reversed? Non‐partisan alternatives in Norwegian local politics | |
| CN108509534B (en) | Personalized music recommendation system based on deep learning and implementation method thereof | |
| WO2019085369A1 (en) | Electronic nose instrument and sensory quality evaluation method for tobacco and tobacco product | |
| CN100421585C (en) | Method for establishing mixed expert system of maintaining cigarette leaf group formulation | |
| Weemstra et al. | Lithological constraints on resource economies shape the mycorrhizal composition of a Bornean rain forest | |
| CN103324758B (en) | A kind of news category method and system | |
| CN104323416A (en) | Flue-cured tobacco leaf formula function discrimination and classified application method | |
| CN110132879A (en) | A kind of flue-cured tobacco note discrimination method based near infrared spectrum | |
| Shott | Toward a theory of the point | |
| ZOBEL | Secondary forest succession in Järvselja, southeastern Estonia: changes in field layer vegetation | |
| CN104323412A (en) | Building and application method for functional tobacco module | |
| CN108414471A (en) | A method of sense organ characterization information is differentiated based near infrared spectrum and sensory evaluating smoking's mutual information | |
| Zagorski-Thomas | Musical meaning and the musicology of record production | |
| CN102339384A (en) | Face recognition method | |
| CN1524461A (en) | Method for establishing mixed expert system of designing cigarette leaf group formulation | |
| CN113936795A (en) | Quantitative analysis method for risk factors of senile lung cancer onset | |
| CN1744115A (en) | Tobacco leaf quality management system and method | |
| CN1975706A (en) | Cigarette organoleptic quality qualitative index estimating method | |
| CN115983692A (en) | Tobacco leaf moistening and sweet flavor evaluation method and device based on random forest algorithm | |
| CN102721792A (en) | Sensory evaluation method for paper-process reconstituted tobacco | |
| Huzaifah et al. | MTCRNN: A multi-scale rnn for directed audio texture synthesis | |
| CN112397156B (en) | Digital flavoring method based on K-means clustering | |
| CN114595365A (en) | Method and device for constructing cigarette feature relevance, electronic equipment and medium | |
| CN113488113B (en) | Industrial use value identification method for redried strip tobacco |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C06 | Publication | ||
| PB01 | Publication | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CX01 | Expiry of patent term | ||
| CX01 | Expiry of patent term |
Granted publication date: 20030723 |