CN109409579A - The method of BP neural network prediction Raw material processing suitability - Google Patents
The method of BP neural network prediction Raw material processing suitability Download PDFInfo
- Publication number
- CN109409579A CN109409579A CN201811147405.2A CN201811147405A CN109409579A CN 109409579 A CN109409579 A CN 109409579A CN 201811147405 A CN201811147405 A CN 201811147405A CN 109409579 A CN109409579 A CN 109409579A
- Authority
- CN
- China
- Prior art keywords
- raw material
- apple
- index
- value
- neural network
- 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.)
- Pending
Links
- 239000002994 raw material Substances 0.000 title claims abstract description 118
- 238000012545 processing Methods 0.000 title claims abstract description 74
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 62
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 51
- 210000005036 nerve Anatomy 0.000 claims abstract description 26
- 241001269238 Data Species 0.000 claims abstract description 12
- 238000005516 engineering process Methods 0.000 claims abstract description 12
- 239000002253 acid Substances 0.000 claims description 43
- 235000013399 edible fruits Nutrition 0.000 claims description 36
- 235000013305 food Nutrition 0.000 claims description 18
- 238000012216 screening Methods 0.000 claims description 17
- 238000010219 correlation analysis Methods 0.000 claims description 16
- 239000000835 fiber Substances 0.000 claims description 14
- 235000011389 fruit/vegetable juice Nutrition 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 6
- FYYHWMGAXLPEAU-UHFFFAOYSA-N Magnesium Chemical compound [Mg] FYYHWMGAXLPEAU-UHFFFAOYSA-N 0.000 claims description 5
- 238000001514 detection method Methods 0.000 claims description 5
- 239000011777 magnesium Substances 0.000 claims description 5
- 229910052749 magnesium Inorganic materials 0.000 claims description 5
- 235000010987 pectin Nutrition 0.000 claims description 5
- 229920001277 pectin Polymers 0.000 claims description 5
- 239000001814 pectin Substances 0.000 claims description 5
- 108090000623 proteins and genes Proteins 0.000 claims description 5
- 239000000126 substance Substances 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- OYPRJOBELJOOCE-UHFFFAOYSA-N Calcium Chemical compound [Ca] OYPRJOBELJOOCE-UHFFFAOYSA-N 0.000 claims description 3
- 102000030523 Catechol oxidase Human genes 0.000 claims description 3
- 108010031396 Catechol oxidase Proteins 0.000 claims description 3
- ZLMJMSJWJFRBEC-UHFFFAOYSA-N Potassium Chemical compound [K] ZLMJMSJWJFRBEC-UHFFFAOYSA-N 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 239000011575 calcium Substances 0.000 claims description 3
- 229910052791 calcium Inorganic materials 0.000 claims description 3
- 235000019784 crude fat Nutrition 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 150000002989 phenols Chemical class 0.000 claims description 3
- 239000011591 potassium Substances 0.000 claims description 3
- 229910052700 potassium Inorganic materials 0.000 claims description 3
- 102000004169 proteins and genes Human genes 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 abstract description 2
- 241000220225 Malus Species 0.000 description 255
- 235000019987 cider Nutrition 0.000 description 69
- 235000015203 fruit juice Nutrition 0.000 description 25
- 238000011156 evaluation Methods 0.000 description 22
- 235000019750 Crude protein Nutrition 0.000 description 19
- 238000004519 manufacturing process Methods 0.000 description 18
- 239000007787 solid Substances 0.000 description 15
- 239000011159 matrix material Substances 0.000 description 14
- 241000607479 Yersinia pestis Species 0.000 description 11
- 201000010099 disease Diseases 0.000 description 11
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 11
- ISWSIDIOOBJBQZ-UHFFFAOYSA-N Phenol Chemical compound OC1=CC=CC=C1 ISWSIDIOOBJBQZ-UHFFFAOYSA-N 0.000 description 10
- 230000006870 function Effects 0.000 description 8
- 238000005070 sampling Methods 0.000 description 8
- 230000001953 sensory effect Effects 0.000 description 8
- 210000000697 sensory organ Anatomy 0.000 description 8
- 239000004575 stone Substances 0.000 description 8
- 235000021022 fresh fruits Nutrition 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 6
- 238000013210 evaluation model Methods 0.000 description 6
- 235000013372 meat Nutrition 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 235000016709 nutrition Nutrition 0.000 description 4
- KRKNYBCHXYNGOX-UHFFFAOYSA-N citric acid Chemical compound OC(=O)CC(O)(C(O)=O)CC(O)=O KRKNYBCHXYNGOX-UHFFFAOYSA-N 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 241000283690 Bos taurus Species 0.000 description 2
- 235000005979 Citrus limon Nutrition 0.000 description 2
- 244000131522 Citrus pyriformis Species 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 235000021466 carotenoid Nutrition 0.000 description 2
- 150000001747 carotenoids Chemical class 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000000556 factor analysis Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013441 quality evaluation Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 244000144730 Amygdalus persica Species 0.000 description 1
- 244000000626 Daucus carota Species 0.000 description 1
- 235000002767 Daucus carota Nutrition 0.000 description 1
- 241000287828 Gallus gallus Species 0.000 description 1
- 235000007688 Lycopersicon esculentum Nutrition 0.000 description 1
- 240000008790 Musa x paradisiaca Species 0.000 description 1
- 235000018290 Musa x paradisiaca Nutrition 0.000 description 1
- 235000006040 Prunus persica var persica Nutrition 0.000 description 1
- 240000003768 Solanum lycopersicum Species 0.000 description 1
- 235000010724 Wisteria floribunda Nutrition 0.000 description 1
- 238000012271 agricultural production Methods 0.000 description 1
- 235000021452 apple slice Nutrition 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- JLQUFIHWVLZVTJ-UHFFFAOYSA-N carbosulfan Chemical compound CCCCN(CCCC)SN(C)C(=O)OC1=CC=CC2=C1OC(C)(C)C2 JLQUFIHWVLZVTJ-UHFFFAOYSA-N 0.000 description 1
- 239000003153 chemical reaction reagent Substances 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000796 flavoring agent Substances 0.000 description 1
- 235000019634 flavors Nutrition 0.000 description 1
- 238000011010 flushing procedure Methods 0.000 description 1
- 235000012055 fruits and vegetables Nutrition 0.000 description 1
- 239000010977 jade Substances 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 235000013622 meat product Nutrition 0.000 description 1
- 235000013336 milk Nutrition 0.000 description 1
- 239000008267 milk Substances 0.000 description 1
- 210000004080 milk Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 235000015277 pork Nutrition 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Evolutionary Computation (AREA)
- Entrepreneurship & Innovation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a kind of methods based on BP neural network prediction Raw material processing suitability, comprising: selects the single index for evaluating the fabricated product of Raw material processing suitability, all multi objectives of relevant raw material are determined according to the single index of fabricated product；It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of raw material, records and arrange historical data formation sample；Using all multiple parameter datas of the raw material of training sample set as input layer, using the single achievement data of the fabricated product of training sample set as output layer, training to artificial nerve network model is stablized；Optimized artificial neural network model；Predict the single index of fabricated product.The present invention can be associated with feedstock specifications and processed goods quality, it is objective qualitatively or quantitatively seek nonlinear correspondence relation complicated between variable while have higher accuracy, may be implemented qualitatively or quantitatively to predict processed goods quality based on feedstock specifications.
Description
Technical field
The present invention relates to manufacture fields.It is more particularly related to which a kind of BP neural network prediction raw material adds
The method of work suitability.
Background technique
Property of raw material is the basis for producing High quality processing product, and therefore, association property of raw material adds with product property, clear raw material
Work suitability, screening high-quality, which process raw material, to be increased income reducing enterprise cost, society and industry development is significant.Raw material is added at present
The research of work suitability mostly uses the methods of step analysis, grey correlation independent analysis material quality or product property, only can be with
Evaluate studied kind processing characteristics, the processing performance of unpredictable unknown sample.Existing a small number of association property of raw material and system
The method of product quality can not handle complicated non-linear correspondence between index merely with linear models such as discriminant function, linear regressions
Relationship, that there are model limitations is big, the scope of application is small, the low problem of prediction unknown sample processing performance accuracy rate.
Artificial neural network is a kind of to be intended to imitate being connected by a large amount of simple computing units for human brain structure and its function
The intelligent information handling system of formation has many advantages, such as MPP, self learning type, adaptivity, is suitable for building
Complicated non-linear correlation model.BP artificial neural network is a kind of Multi-layered Feedforward Networks by Back Propagation Algorithm training,
It is one of current most widely used neural network model, it can be achieved that being input to qualitatively or quantitatively predicting and have higher for output
Accuracy rate.In recent years BP neural network model in processing industry field using increasing, including identify, classify and be classified plus
Work process simulation and control, the prediction of single index value etc., achieve certain effect, but in terms of Raw material processing suitability prediction
Application have not been reported.
Summary of the invention
It is an object of the invention to solve at least the above problems, and provide the advantages of at least will be described later.
It is a still further object of the present invention to provide a kind of method of BP neural network prediction Raw material processing suitability,
Feedstock specifications and processed goods quality can be associated with, qualitatively or quantitatively seek complicated nonlinear correspondence relation between variable objective
While have higher accuracy, may be implemented qualitatively or quantitatively to predict processed goods quality based on feedstock specifications.
In order to realize these purposes and other advantages according to the present invention, provide a kind of pre- based on BP artificial neural network
The method for surveying Raw material processing suitability, comprising:
The single index for evaluating the fabricated product of Raw material processing suitability is selected, according to the single index of fabricated product
Determine all multi objectives of relevant raw material；
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of raw material, records and arrange and go through
History data form sample, the format of sample are as follows: each historical data includes all multiple parameter datas and fabricated product of raw material
Single achievement data；
A certain proportion of historical data is selected to form training sample set, remaining historical data forms detection sample set；
Using all multiple parameter datas of the raw material of training sample set as input layer, with the single of the fabricated product of training sample set
Achievement data is output layer, determines input layer number, hidden layer neuron number and output layer neuron number, is determined
Frequency of training, learning rate, momentum, error amount determine the activation primitive of hidden layer, output layer, construct based on the artificial of BP algorithm
Neural network model, and be trained to artificial nerve network model and stablize；
It is tested with detecting sample set to artificial nerve network model, optimized artificial neural network model；
Using all multi objectives of raw material as input layer, the single of fabricated product is predicted according to the artificial nerve network model of optimization
Index.
Preferably, the historical data of 70-90% is selected to form training sample set, remaining data form detection sample
Collection.
Preferably, sample includes at least 30 historical datas.
Preferably, frequency of training is 500-8000 times, learning rate 0.1-0.5, momentum 0.1-0.8, and error amount is
0.00001-0.1。
Preferably, the index of raw material includes base values, physical and chemical index and processing index, and base values includes raw material
Volume, density, color, physical and chemical index includes the sugar of raw material, acid, protein, phenols, pectin content, polyphenol oxidase activity, thick
Fiber, crude fat, potassium, calcium, magnesium, processing index include edible ratio, fruit shape index, the brown stain degree of raw material.
Preferably, the index of fabricated product be the turbidity of juice, stability, comprehensive score, quality grade wherein
One kind or the color difference a value of dried product, brittleness, comprehensive score, the comprehensive score of quality grade or fresh food, quality grade are wherein
It is a kind of.
Preferably, when determining all multi objectives of relevant raw material according to the single index of fabricated product, sieve artificial first
Feedstock specifications are selected, correlation analysis then is carried out to the single index of each single item feedstock specifications and fabricated product, removal is related
Coefficient is lower than the feedstock specifications of preset correlation coefficient threshold.
Preferably, it is lower than preset error threshold with training sample set training artificial nerve network model to error
Artificial nerve network model is stablized.
When preferably, to detect sample set test artificial nerve network model, compare the single index of fabricated product
Predicted value and measured value meet preset threshold value, then artificial nerve network model is stablized, otherwise this historical data is added to
Training sample set is trained, optimized artificial neural network model.
Preferably, with detect sample set test artificial nerve network model to fabricated product single index predicted value
Being higher than preset correlation coefficient threshold lower than preset error threshold or related coefficient with the error of measured value is artificial neuron
Network model is stablized.
The present invention is include at least the following beneficial effects:
The first, the present invention is associated with former under conditions of with a certain amount of learning sample using BP artificial nerve network model
Expect index and processed goods quality, be not necessarily to empirical equation and mathematical model, it is objective qualitatively or quantitatively seek it is complicated between variable
There is higher accuracy while nonlinear correspondence relation, it is qualitative or quantitative objective, accurately based on feedstock specifications to may be implemented
Predict fabricated product quality；
The second, the training of the single achievement data of all multi objective-fabricated products of a large amount of different raw materials of present invention progress
And test, the single index of BP artificial nerve network model Accurate Prediction fabricated product is constructed, target system is improved from source
The quality of product improves the associated reasonability of raw material and fabricated product, can be greatly promoted the development of industry.
Further advantage, target and feature of the invention will be partially reflected by the following instructions, and part will also be by this
The research and practice of invention and be understood by the person skilled in the art.
Specific embodiment
Below with reference to example, the present invention is described in further detail, to enable those skilled in the art referring to specification text
Word can be implemented accordingly.
It should be appreciated that such as " having ", "comprising" and " comprising " term used herein do not allot one or more
The presence or addition of a other elements or combinations thereof.It should be noted that experimental method described in following embodiments, such as without spy
Different explanation, is conventional method, the reagent and material commercially obtain unless otherwise specified.
The present invention provides a kind of method based on BP neural network prediction Raw material processing suitability, comprising:
The single index for evaluating the fabricated product of Raw material processing suitability is selected, according to the single index of fabricated product
Determine all multi objectives of relevant raw material, the index of raw material should be main producing region principal item, and index value answers difference big, the finger of raw material
Mark includes base values, physical and chemical index and processing index, and base values includes the volume, density, color of raw material, physical and chemical index packet
Sugar, acid, protein, phenols, pectin content, polyphenol oxidase activity, crude fibre, crude fat, potassium, calcium, the magnesium of raw material are included, is processed
Index includes edible ratio, fruit shape index, the brown stain degree of raw material；The index of fabricated product is the turbidity of juice, stability, comprehensive
Close the comprehensive of scoring, the color difference a value of quality grade one of which or dried product, brittleness, comprehensive score, quality grade or fresh food
Scoring, quality grade one of which are closed, artificial screening goes out feedstock specifications first, then to each single item feedstock specifications and fabricated product
Single index carry out correlation analysis, removal related coefficient be lower than preset correlation coefficient threshold feedstock specifications, screening most
Whole raw material core index；
All multi objectives (core index) with raw material are that variable, processing technology and parameter are quantitative progress Raw material processing, note
It records and arranges historical data and form sample, sample includes at least 30 historical datas, the format of sample are as follows: each historical data
The single achievement data of all multiple parameter datas and fabricated product including raw material；
A certain proportion of historical data (preferably 70-90%) is selected to form training sample set, remaining historical data is (preferably
15-30%) form detection sample set；
Using all multiple parameter datas of the raw material of training sample set as input layer, with the single of the fabricated product of training sample set
Achievement data is output layer, determines input layer number, hidden layer neuron number and output layer neuron number, is determined
Frequency of training, learning rate, momentum, error amount determine the activation primitive of hidden layer, output layer, and hidden layer neuron number is by counting
Software for calculation generates according to statistics, and frequency of training is 500-8000 times, learning rate 0.1-0.5, momentum 0.1-0.8, error amount
For 0.00001-0.1, the artificial nerve network model based on BP algorithm is constructed, and carries out training of human artificial neural networks model to accidentally
Difference is that artificial nerve network model is stablized lower than preset error threshold；
Whether neural metwork training is completed to commonly use error function (also referred to as objective function) E to measure, when error function is less than
Stop the training of neural network when the value of some setting, error function is to measure reality output vector YkWith expected value vector
TkThe function of error size multiplies error function frequently with two to be defined as(or)k
=1,2 ..., N is training sample number；
It is tested with detecting sample set to artificial nerve network model, compares the predicted value of the single index of fabricated product
And measured value, with detect sample set test artificial nerve network model to fabricated product single index predicted value and measured value
Error to be higher than preset correlation coefficient threshold lower than preset error threshold or related coefficient be artificial nerve network model
Stablize, for example, for verifying model stability cross validation (no less than 3 times) can be carried out, if detecting sample after obtained cross validation
The coefficient of determination is all larger than 0.9 after the predicted value and actual value linear fit of this collection, otherwise this historical data is added to training
Sample set is trained, optimized artificial neural network model；
Using all multi objectives of raw material as input layer, the single of fabricated product is predicted according to the artificial nerve network model of optimization
Index.
Below using apple as raw material, selected apple raw material is the corresponding main cultivation apple type in apple main product each province, and
According to it is precocious, in it is ripe, late-maturing apple raw material is divided into three classes, specifically have: it is precocious: the tai shan rosy clouds of dawn, Shaanxi Qin Yang etc., in
It is ripe: Shaanxi Huang marshal, Henan China jade etc., it is late-maturing: the long No. two/Hua Fu of richness in Liaoning, Shandong state light, Gansu Qin Guan, Gansu flower ox, newly
Boundary Fuji, Ningxia Qiao Najin etc., the processing suitability for carrying out apple predicted, the apple sample of each example cover it is precocious,
In ripe, late variety.
<example 1>
Selecting the cider turbidity for evaluating apple processing suitability is single index, comprising the following steps:
(1) apple sample is selected
41 apple varieties from various parts of the country are chosen as experimental raw, part variety name, the place of production are shown in Table 1, fruit
Real maturity period sampling, has no mechanical damage, no disease and pests harm；
1 part apple variety title of table and the place of production
(2) apple feedstock specifications are determined, and obtain apple feedstock specifications data and cider haze numbers evidence；
(3) apple raw material core index is screened；
The correlativity for establishing apple feedstock specifications data Yu cider haze numbers evidence, to cider haze numbers according to
Apple feedstock specifications data carry out correlation analysis, and the results are shown in Table 2, obtains core size, pulp L value, pericarp a value and lemon
Core index of lemon 4 indexs of acid as apple raw material；
The correlativity of table 2 apple feedstock specifications data and cider haze numbers evidence
(4) neural-network learning model is constructed
35 apple samples of random screening establish learning model, residue 6 as training sample from 41 apple samples
A apple sample carries out the prediction of cider turbidity as forecast sample, and mode input layer is core size, pulp L value, pericarp a
Value and 4 raw material core index data of citric acid, model output layer are the corresponding cider haze values of Apple, input layer
Lowest level is model-aided layer, is added automatically by software, and the optimal implicit number of plies of model is 5, remaining each training parameter selection is such as
Under: maximum cycle 4000, learning rate 0.2, factor of momentum 0.2, error amount 0.001；
(5) forecast sample is verified
Using neural-network learning model, the cider turbidity of 6 apple samples is predicted, as a result such as 3 institute of table
Show, relative error is less than 3%, and preset threshold value is that less than 8%, 6 apple sample prediction of absolute relative error is accurate, prediction
Accuracy rate is 100%.
3 neural network prediction result of table
<example 2>
Selecting the cider comprehensive score for evaluating apple processing suitability is single index, comprising the following steps:
(1) apple sample is selected
33 apple varieties from various parts of the country are chosen as experimental raw, part apple variety title, the place of production are shown in Table
4, fructescence sampling has no mechanical damage, no disease and pests harm；
4 part apple variety title of table and the place of production
(2) evaluation method of cider comprehensive score is determined
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of apple, records and arrange and go through
History data, measure the data of cider, cider data include soluble solid, titratable acid, crude fibre, crude protein, Vc,
Reduced sugar, total sugar content, fruit juice L value, fruit juice a value, fruit juice b value, crushing juice rate, turbidity；
Factor analysis is utilized to above-mentioned 13 indexs of cider, setting output absolute value data is greater than 0.5 data,
It the results are shown in Table shown in 5；
5 cider index factor of table analysis rotation component matrix
Note: PC1-PC5 respectively indicates the 1st to the 5th main gene
According to table 5 it is found that screening representative index of the higher index of weighted value as each factor in each factor, therefore sieve
Select total phenol content, L value, titratable acid, turbidity and Vc4 are fruit juice core index, and record each cider core index pair
The cider core index data answered；
Y-P judgment matrix is established with analytic hierarchy process (AHP), as shown in table 6, is obtained to after matrix characteristic vector normalized
To cider core index weight；
6 judgment matrix Y-P of table
Fruit juice comprehensive evaluation model is established according to table 6 and then obtains fruit juice comprehensive score, the corresponding fruit juice of each cider
Comprehensive score are as follows: Y (comprehensive score)=total phenol content × 0.416+L value × 0.027+ titratable acid content × 0.164+ turbidity
Content × 0.060 0.092+Vc；
(3) apple feedstock specifications are determined, and obtain apple feedstock specifications data and cider comprehensive score data；
(4) apple raw material core index is screened；
The correlativity for establishing apple feedstock specifications data Yu cider comprehensive score data, to cider comprehensive score number
Correlation analysis is carried out according to apple feedstock specifications data, the results are shown in Table 7, obtains density, fruit stone ratio, pulp L value, fruit
10 skin hardness, flesh firmness, pH, titratable acid, soluble solid, moisture content, crude protein indexs are as apple raw material
Core index；
The correlativity of table 7 apple feedstock specifications data and cider comprehensive score data
(5) neural-network learning model is constructed
24 apple samples of random screening establish learning model, residue 9 as training sample from 33 apple samples
A apple sample carries out cider comprehensive score number it was predicted that mode input layer is density, fruit stone ratio, fruit as forecast sample
10 meat L value, Rind hardness, flesh firmness, pH, titratable acid, soluble solid, moisture content, crude protein raw material cores refer to
Data are marked, model output layer is cider comprehensive score, and using data statistics software for calculation, the optimal implicit number of plies of model is by software
It automatically generates, wherein frequency of training is 1000 times, learning rate 0.1, momentum 0.2, error amount 0.001；
(6) forecast sample is verified
Using neural-network learning model, the cider comprehensive score of 9 apple samples is predicted, as a result such as table 8
Shown, preset threshold value is that less than 8%, 8 apple sample prediction of absolute relative error is accurate, and predictablity rate is
88.89%.
8 neural network prediction result of table
The corresponding sample set data for not meeting preset threshold are added to training sample set to be trained, optimize artificial mind
Through network model, the new artificial neural network learning model of framework.
<example 3>
Selecting the cider comprehensive score for evaluating apple processing suitability is single index, comprising the following steps:
(1) apple sample is selected
30 apple varieties from various parts of the country are chosen as experimental raw, part apple variety title, the place of production are shown in Table
9, fructescence sampling has no mechanical damage, no disease and pests harm；
9 part apple variety title of table and the place of production
(2) evaluation method of cider comprehensive score is determined
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of apple, records and arrange and go through
History data carry out sensory evaluation scores to cider, and standards of grading are as shown in table 10, and appraisal result is as shown in table 11；
10 cider sense organ evaluating meter of table
The sense organ overall score of 11 cider of table
(3) apple feedstock specifications are determined, and obtain apple feedstock specifications data and cider comprehensive score data；
(4) apple raw material core index is screened；
The correlativity for establishing apple feedstock specifications data Yu cider comprehensive score data, to cider comprehensive score number
As a result as shown in table 12 pericarp L value, pulp L value, pH value 3 are obtained according to correlation analysis is carried out with apple feedstock specifications data
Core index of the index as apple raw material；
12 feedstock specifications of table and cider sense organ overall score correlation analysis
(5) neural-network learning model is constructed
25 apple samples of random screening establish learning model, residue 5 as training sample from 30 apple samples
A apple sample as forecast sample carry out cider comprehensive score number it was predicted that mode input layer be pericarp L value, pulp L value,
3 raw material core index data of pH value, model output layer are cider comprehensive score, utilize data statistics software for calculation, model
The optimal implicit number of plies is automatically generated by software, wherein frequency of training is 3000 times, learning rate 0.3, momentum 0.2, error amount
It is 0.001；
(6) forecast sample is verified
Using neural-network learning model, the cider comprehensive score of 5 apple samples is predicted, as a result such as table 13
Shown, relative error is less than 6%, and preset threshold value is that less than 8%, 5 apple sample prediction of absolute relative error is accurate, in advance
Surveying accuracy rate is 100%.
13 neural network prediction result of table
<example 4>
Selecting the cider quality grade for evaluating apple processing suitability is single index, comprising the following steps:
(1) apple sample is selected
33 apple varieties from various parts of the country are chosen as experimental raw, part apple variety title, the place of production are shown in Table
14, fructescence sampling has no mechanical damage, no disease and pests harm；
14 part apple variety title of table and the place of production
(2) evaluation method of cider quality grade is determined
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of apple, records and arrange and go through
History data, measure the data of cider, cider data include soluble solid, titratable acid, crude fibre, crude protein, Vc,
Reduced sugar, total sugar content, fruit juice L value, fruit juice a value, fruit juice b value, crushing juice rate, turbidity；
Factor analysis is utilized to above-mentioned 13 indexs of cider, setting output absolute value data is greater than 0.5 data,
As a result with table 5；
According to table 5 it is found that screening representative index of the higher index of weighted value as each factor in each factor, therefore sieve
Selecting total phenol content, L value, titratable acid, turbidity and Vc is cider core index, and it is corresponding to record each cider core index
Fruit juice core index data；
Y-P judgment matrix is established with analytic hierarchy process (AHP), with table 6, to obtaining apple after matrix characteristic vector normalized
Fruit juice core index weight；
Fruit juice comprehensive evaluation model is established according to table 6 and then obtains fruit juice comprehensive score, the corresponding apple of each cider
Juice comprehensive score are as follows: Y (comprehensive score)=total phenol content × 0.416+L value × 0.027+ titratable acid content × 0.164+ turbidity
Content × 0.060 0.092+Vc；
Corresponding apple raw material is divided into 5 quality grades, and the apple according to the corresponding fruit juice comprehensive score of each cider
The grade probability that fruit raw material corresponds to its quality grade is 1, and the grade probability of remaining quality grade is 0, specifically: cider is comprehensive
Scoring >=0.8 is level-one, and 0.7≤cider comprehensive score < 0.8 is second level, and 0.6≤cider comprehensive score < 0.7 is three
Grade, 0.5≤cider comprehensive score < 0.6 are level Four, and cider comprehensive score < 0.5 is Pyatyi, the results are shown in Table 15；
15 apple raw material of table corresponds to cider comprehensive score and quality grade
(3) apple feedstock specifications are determined, and obtain apple feedstock specifications data and cider quality grade data；
(4) apple raw material core index is screened；
The correlativity for establishing apple feedstock specifications data Yu cider quality grade data, to cider quality grade number
It is hard to obtain density, fruit stone ratio, pulp L value, pericarp as a result with table 7 according to correlation analysis is carried out with apple feedstock specifications data
The core of degree, flesh firmness, pH, titratable acid, soluble solid, 10 moisture content, crude protein indexs as apple raw material
Index；
(5) neural-network learning model is constructed
24 apple samples of random screening establish learning model, residue 9 as training sample from 33 apple samples
A apple sample carries out cider quality grade number it was predicted that mode input layer is density, fruit stone ratio, fruit as forecast sample
10 meat L value, Rind hardness, flesh firmness, pH, titratable acid, soluble solid, moisture content, crude protein raw material cores refer to
Data are marked, model output layer is cider quality grade, and using data statistics software for calculation, the optimal implicit number of plies of model is by software
It automatically generates, wherein frequency of training is 1000 times, learning rate 0.1, momentum 0.1, error amount 0.001；
(6) forecast sample is verified
Using neural-network learning model, the cider quality grade of 9 apple samples is predicted, as a result such as table 16
Shown, 9 apple sample predictions are accurate, predictablity rate 100%.
16 neural network prediction result of table
<example 5>
Selecting the cider quality grade for evaluating apple processing suitability is single index, comprising the following steps:
(1) apple sample is selected
30 apple varieties from various parts of the country are chosen as experimental raw, are had no mechanical damage, no disease and pests harm；
17 part apple variety title of table and the place of production
(2) evaluation method of cider quality grade is determined
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of apple, records and arrange and go through
History data measure the fruit juice L value of cider, and corresponding apple raw material is divided into 3 product according to the corresponding fruit juice L value of each cider
Matter grade, and it is 1 that the apple raw material, which corresponds to the grade probability of its quality grade, the grade probability of remaining quality grade is 0, specifically
Are as follows: 65.00 value≤70.00 fruit juice L < be level-one, 60.00 value≤65.00 fruit juice L < be second level, 55.00 < fruit juice L values≤
60.00 be three-level, is specifically shown in Table 18；
18 fruit juice L value of table and corresponding cider quality grade
(3) apple raw material core index is screened；
The correlativity for establishing apple feedstock specifications data Yu cider L value, refers to cider L Value Data and apple raw material
It marks data and carries out correlation analysis, as a result as shown in table 19, obtain fruit shape index, pulp L value, pH value and carotenoid 4
Core index of the index as apple raw material；
19 feedstock specifications of table and fruit juice L value correlation analysis
(5) neural-network learning model is constructed
25 apple samples of random screening establish learning model, residue 5 as training sample from 30 apple samples
A apple sample carries out cider quality grade number it was predicted that mode input layer is fruit shape index, pulp L as forecast sample
4 value, pH value and carotenoid raw material core index data, model output layer are cider quality grade, are united using data
Software for calculation is counted, the optimal implicit number of plies of model is automatically generated by software, wherein frequency of training is 3000 times, learning rate 0.3,
Momentum is 0.2, error amount 0.001；
(6) forecast sample is verified
Using neural-network learning model, the cider quality grade of 5 apple samples is predicted, as a result such as table 20
Shown, 5 apple sample predictions are accurate, predictablity rate 100%.
20 neural network prediction result of table
<example 6>
Selecting the dry apple comprehensive score for evaluating apple processing suitability is single index, including following step
It is rapid:
(1) apple sample is selected
34 apple varieties from various parts of the country are chosen as experimental raw, part apple variety title, the place of production are shown in Table
21, fructescence sampling has no mechanical damage, no disease and pests harm；
21 part apple variety title of table and the place of production
(2) evaluation method of dry apple comprehensive score is determined
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of apple, records and arrange and go through
History data measure the data of dry apple；
Dried product integrated quality mainly considers that its organoleptic quality, processing quality and nutritional quality, organoleptic quality include color
Quality (L, a, b value), texture quality (hardness, brittleness), puffed degree, processing quality include yield rate, rehydration ratio, moisture content,
Nutritional quality includes soluble solid, titratable acid, total phenol content, pectin content, soluble sugar, sugar-acid ratio, crude protein, thick
Fiber amounts to 17 indexs (initial dried product evaluation index), carries out dimensionality reduction (factorial analysis) to 17 indexs of dried product, can obtain
To 6 factors, the 77% of all information can be accounted for, the results are shown in Table 22, the characteristic value of preceding 6 factors is greater than 1, adds up variance contribution ratio
It is 77.402%, the factor 1 mainly combines the information of titratable acid and sugar-acid ratio, and the two shows extremely significant correlation (R=-
0.849), and the weighted value of titratable acid is higher, therefore screens representative index of the titratable acid as the factor 1, and the factor 2 is main
The information of b value, brittleness and total phenol is combined, wherein brittleness is significant related (R=0.426) to total phenol content, with the extremely significant phase of b value
Close (R=-0.472), and brittleness is the important organoleptic indicator of measuring apple dried product processing quality, therefore screen brittleness as because
The representative index of son 2, the factor 4 mainly combine the information of L value, a value, and L value represents bright-dark degree, and a value represents red green degree,
Dried product Color Quality is embodied, and shows extremely significant correlation (R=-0.865), screens the higher L value of weighted value herein
As the representative index of the factor 4, meanwhile, puffed degree, the weight of soluble sugar and crude protein in the factor 3, the factor 5 and the factor 6
Value is apparently higher than other indexs, therefore screens puffed degree, soluble sugar and crude protein respectively as the factor 3, the factor 5 and the factor 6
Representative index titratable acid, brittleness, puffed degree, L value, soluble sugar and thick egg are to sum up filtered out in 17 index of quality
It is white to be used as dried product evaluation index；
22 dried product index factor of table analysis rotation component matrix
Note: PC1-PC6 respectively indicates the 1st to the 6th main gene
Y-P judgment matrix is established with analytic hierarchy process (AHP), as shown in table 23, consistency ratio (consistency
Ratio, CR) it is 0.03, less than 0.1, it is believed that judgment matrix approach is acceptable, obtains to after matrix characteristic vector normalized
To dry apple core index weight；
23 judgment matrix Y-P of table
Dry apple comprehensive evaluation model is established according to table 23 and then obtains fruit juice comprehensive score, each dry apple
Corresponding dry apple comprehensive score are as follows: Y (comprehensive score)=L value × 0.3724+ brittleness × 0.2665+ puffed degree ×
0.1583+ titratable acid content × 0.0890+ soluble sugar content × crude protein content × 0.0569 0.0569+；
According to the dry apple quality comprehensive evaluation model of building, it is comprehensive to calculate 34 dry apple sample qualities
Point, it the results are shown in Table 24；
The sequence of 24 dried product integrated quality of table and comprehensive score
(3) apple feedstock specifications are determined, and obtain apple feedstock specifications data and dry apple comprehensive score data；
(4) apple raw material core index is screened；
The correlativity of apple feedstock specifications data Yu dry apple comprehensive score data is established, it is comprehensive to dry apple
It closes score data and apple feedstock specifications data carries out correlation analysis and as a result as shown in Table 25 obtain fruit shape index, pulp a
Value, pH value, titratable acid content, VCContent, fruit stone ratio, crude protein content, pulp b value, density, soluble solid contain
The core index of amount, 12 crude fiber content, total sugar content indexs as apple raw material；
The correlativity of table 25 apple feedstock specifications data and cider comprehensive score data
(5) neural-network learning model is constructed
29 apple samples of random screening establish learning model, residue 5 as training sample from 34 apple samples
A apple sample carries out dry apple comprehensive score number it was predicted that mode input layer is fruit shape index, fruit as forecast sample
Meat a value, pH value, titratable acid content, VCContent, fruit stone ratio, crude protein content, pulp b value, density, soluble solid
12 content, crude fiber content, total sugar content raw material core index data, model output layer are dry apple comprehensive score,
The optimal implicit number of plies of model is 9, maximum cycle 2000, learning rate 0.3, factor of momentum 0.2, error amount 0.001；
(6) forecast sample is verified
Using neural-network learning model, the dry apple comprehensive score of 5 apple samples is predicted, as a result such as
Shown in table 26, preset threshold value is that less than 8%, 4 apple sample prediction of absolute relative error is accurate, and predictablity rate is
80%.
26 neural network prediction result of table
The corresponding sample set data for not meeting preset threshold are added to training sample set to be trained, optimize artificial mind
Through network model, the new artificial neural network learning model of framework.
<example 7>
Selecting the dry apple comprehensive score for evaluating apple processing suitability is single index, including following step
It is rapid:
(1) apple sample is selected
34 apple varieties from various parts of the country are chosen as experimental raw, part apple variety title, the place of production are shown in Table
27, fructescence sampling has no mechanical damage, no disease and pests harm；
27 part apple variety title of table and the place of production
(2) evaluation method of dry apple comprehensive score is determined
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of apple, records and arrange and go through
History data carry out sensory evaluation scores to dry apple, and standards of grading are as shown in table 28, and appraisal result is as shown in table 29；
28 dry apple sense organ evaluating meter of table
29 dry apple sense organ overall score of table
(3) apple feedstock specifications are determined, and obtain apple feedstock specifications data and dry apple comprehensive score data；
(4) apple raw material core index is screened；
The correlativity of apple feedstock specifications data Yu dry apple comprehensive score data is established, it is comprehensive to dry apple
It closes score data and apple feedstock specifications data carries out correlation analysis and as a result as shown in table 30 obtain pericarp L value, pulp a
10 value, pulp b value, pH, titratable acid, soluble solid, crude fibre, Vc, reduced sugar, magnesium indexs are as apple raw material
Core index；
30 feedstock specifications of table and dry apple sense organ overall score correlation analysis
(5) neural-network learning model is constructed
28 apple samples of random screening establish learning model, residue 6 as training sample from 34 apple samples
A apple sample carries out dry apple comprehensive score number it was predicted that mode input layer is pericarp L value, pulp as forecast sample
10 a value, pulp b value, pH, titratable acid, soluble solid, crude fibre, Vc, reduced sugar, magnesium raw material core index data,
Model output layer is dry apple comprehensive score, and the optimal hidden layer number of plies of model is 9, maximum cycle 8000, study
Rate 0.2, factor of momentum 0.1, error amount 0.01；
(6) forecast sample is verified
Using neural-network learning model, the dry apple comprehensive score of 6 apple samples is predicted, as a result such as
Shown in table 31, for relative error less than 4%, preset threshold value is that less than 8%, 6 apple sample prediction of absolute relative error is quasi-
Really, predictablity rate 100%.
31 neural network prediction result of table
<example 8>
Selecting the dry apple quality grade for evaluating apple processing suitability is single index, including following step
It is rapid:
(1) apple sample is selected
34 apple varieties from various parts of the country are chosen as experimental raw, part apple variety title, the place of production such as table
Shown in 32, fructescence sampling has no mechanical damage, no disease and pests harm；
32 part apple variety title of table and the place of production
(2) evaluation method of dry apple quality grade is determined
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of apple, records and arrange and go through
History data, measure the data of dried product, dried product data include soluble solid, titratable acid, crude fibre, crude protein, Vc,
Reduced sugar, total sugar content, dried product L value, dried product a value, dried product b value, crushing juice rate, turbidity；
Dried product integrated quality mainly considers that its organoleptic quality, processing quality and nutritional quality, organoleptic quality include color
Quality (L, a, b value), texture quality (hardness, brittleness), puffed degree, processing quality include yield rate, rehydration ratio, moisture content,
Nutritional quality includes soluble solid, titratable acid, total phenol content, pectin content, soluble sugar, sugar-acid ratio, crude protein, thick
Fiber amounts to 17 indexs (initial dried product evaluation index), carries out dimensionality reduction (factorial analysis) to 17 indexs of dried product, can obtain
To 6 factors, the 77% of all information can be accounted for, as a result with table 22, the characteristic value of preceding 6 factors is greater than 1, adds up variance contribution ratio
It is 77.402%, the factor 1 mainly combines the information of titratable acid and sugar-acid ratio, and the two shows extremely significant correlation (R=-
0.849), and the weighted value of titratable acid is higher, therefore screens representative index of the titratable acid as the factor 1, and the factor 2 is main
The information of b value, brittleness and total phenol is combined, wherein brittleness is significant related (R=0.426) to total phenol content, with the extremely significant phase of b value
Close (R=-0.472), and brittleness is the important organoleptic indicator of measuring apple dried product processing quality, therefore screen brittleness as because
The representative index of son 2, the factor 4 mainly combine the information of L value, a value, and L value represents bright-dark degree, and a value represents red green degree,
Dried product Color Quality is embodied, and shows extremely significant correlation (R=-0.865), screens the higher L value of weighted value herein
As the representative index of the factor 4, meanwhile, puffed degree, the weight of soluble sugar and crude protein in the factor 3, the factor 5 and the factor 6
Value is apparently higher than other indexs, therefore screens puffed degree, soluble sugar and crude protein respectively as the factor 3, the factor 5 and the factor 6
Representative index titratable acid, brittleness, puffed degree, L value, soluble sugar and thick egg are to sum up filtered out in 17 index of quality
It is white to be used as dried product evaluation index；
Y-P judgment matrix is established with analytic hierarchy process (AHP), with table 23, consistency ratio (consistency ratio, CR)
It is 0.03, less than 0.1, it is believed that judgment matrix approach is acceptable, to obtaining dried apple slices after matrix characteristic vector normalized
Product core index weight；
Dry apple comprehensive evaluation model is established according to table 23 and then obtains fruit juice comprehensive score, each dry apple
Corresponding dry apple comprehensive score are as follows: Y (comprehensive score)=L value × 0.3724+ brittleness × 0.2665+ puffed degree ×
0.1583+ titratable acid content × 0.0890+ soluble sugar content × crude protein content × 0.0569 0.0569+；
According to the dry apple quality comprehensive evaluation model of building, it is comprehensive to calculate 34 dry apple sample qualities
Point, and it is classified, score 0.7-0.8 is level-one, and 0.6-0.7 is second level, and 0.5-0.6 is three-level, 0.4-0.5 tetra-
Grade is Pyatyi less than 0.4, the results are shown in Table 33；
The sequence of 33 dried product integrated quality of table and scoring rank
(3) apple feedstock specifications are determined, and obtain apple feedstock specifications data and dry apple quality grade data；
(4) apple raw material core index is screened；
The correlativity of apple feedstock specifications data Yu dry apple comprehensive score data is established, it is comprehensive to dry apple
It closes score data and apple feedstock specifications data carries out correlation analysis and obtain fruit shape index, pulp a value, pH as a result with table 25
Value, titratable acid content, VCContent, fruit stone ratio, crude protein content, pulp b value, density, soluble solid content, thick fibre
The core index of 12 dimension hplc, total sugar content indexs as apple raw material；
(5) neural-network learning model is constructed
29 apple samples of random screening establish learning model, residue 5 as training sample from 34 apple samples
A apple sample carries out dry apple quality grade number it was predicted that mode input layer is fruit shape index, fruit as forecast sample
Meat a value, pH value, titratable acid content, VCContent, fruit stone ratio, crude protein content, pulp b value, density, soluble solid
12 content, crude fiber content, total sugar content raw material core index data, model output layer are dry apple quality grade,
The optimal implicit number of plies of model is automatically generated by software, and the number of plies is 9, maximum cycle 2000, learning rate 0.3, factor of momentum
0.2, error amount 0.001；
(6) forecast sample is verified
Using neural-network learning model, the dry apple comprehensive score of 5 apple samples is predicted, as a result such as
Shown in table 34,5 apple sample predictions are accurate, predictablity rate 100%.
34 neural network prediction result of table
<example 9>
Selecting the dry apple quality grade for evaluating apple processing suitability is single index, including following step
It is rapid:
(1) apple sample is selected
34 apple varieties from various parts of the country are chosen as experimental raw, wherein it is precocious, in ripe and late variety
Each 10, part apple variety title, the place of production are shown in Table 35, and fructescence sampling has no mechanical damage, no disease and pests harm；
35 part apple variety title of table and the place of production
(2) evaluation method of dry apple quality grade is determined
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of apple, records and arrange and go through
History data measure the dried product L value of dried product, select the L value index of dried product to react dried product Color Quality, according to each
Corresponding apple raw material is divided into 3 quality grades by the corresponding dried product L value of dried product, and L value 75.00-85.00 is level-one,
65.00-75.00 is second level, and 54.00-65.00 is three-level, is specifically shown in Table 36；
36 dried product L value of table and corresponding product quality grade
(3) apple raw material core index is screened；
The correlativity for establishing apple feedstock specifications data Yu dry apple L value, to dry apple dry apple L
Value Data and apple feedstock specifications data carry out correlation analysis and as a result as shown in table 37 obtain fruit shape index, pulp a value, pH
Value, titratable acid content, VCCore index of 5 indexs of content as apple raw material；
37 feedstock specifications of table and dried product L value correlation analysis
(5) neural-network learning model is constructed
29 apple samples of random screening establish learning model, residue 5 as training sample from 34 apple samples
A apple sample carries out dry apple quality grade number it was predicted that mode input layer is fruit shape index, fruit as forecast sample
Meat a value, pH value, titratable acid content, VC5 raw material core index data of content, model output layer are dry apple quality
Grade, the optimal implicit number of plies of model are 6, maximum cycle 5000, learning rate 0.2, factor of momentum 0.2, error amount
0.00001；
(6) forecast sample is verified
Using neural-network learning model, the dry apple quality grade of 5 apple samples is predicted, as a result such as
Shown in table 38,5 apple sample predictions are accurate, predictablity rate 100%.
38 neural network prediction result of table
<example 9>
Selecting the apple fresh food organoleptic quality for evaluating apple processing suitability is single index, comprising the following steps:
(1) apple sample is selected
40 apple varieties from various parts of the country are chosen as experimental raw, fructescence samples, have no mechanical damage,
No disease and pests harm；
(2) evaluation method of apple fresh food organoleptic quality is determined
Determine that the sensory evaluation index for evaluating apple fresh food sensory evaluation scores is fruit size, fruit face color, pulp matter
How much are ground, flavor and juice, establish the corresponding relationship of sensory evaluation index Yu fruit organoleptic quality, specific as shown in table 39, really
Determine fruit sensory evaluation scores, it is specific as shown in table 40；
Table 39 eats sensory evaluation index and corresponding representative fraction raw
Note: 75%, 50%, 25% refers to the flushing color coverage rate of fruit
40 sensory evaluation score of table
(3) apple fresh fruit Instrument measuring index is measured；
Determine that fresh fruit Instrument measuring index is volume, pericarp L value, pericarp a value, fruit according to fresh food organoleptic quality evaluations index
Skin b value, flesh firmness, titratable acid, soluble sugar, crushing juice rate measure the corresponding Instrument measuring achievement data of each apple fresh fruit；
(4) neural-network learning model is constructed
34 apple samples of random screening establish learning model, residue 6 as training sample from 40 apple samples
A apple sample as forecast sample carry out apple fresh food organoleptic quality number it was predicted that mode input layer be volume, pericarp L value,
Pericarp a value, pericarp b value, flesh firmness, titratable acid, soluble sugar, 8 fresh fruit Instrument measuring achievement datas of crushing juice rate, model
Output layer is that apple eats organoleptic quality raw, and the optimal implicit number of plies of model is 6, and maximum cycle 3000, learning rate 0.2 moves
Measure the factor 0.2, error amount 0.001；
(5) forecast sample is verified
Using neural-network learning model, the apple fresh food organoleptic quality grade of 6 apple samples is predicted, as a result
As shown in table 41, for 6 apple sample relative errors less than 3%, preset threshold value is less than 3%, 6 apple of absolute relative error
Fruit sample predictions are accurate, predictablity rate 100%.
41 neural network prediction result of table
<example 10>
Selecting the apple fresh food organoleptic quality grade for evaluating apple processing suitability is single index, including following step
It is rapid:
(1) apple sample is selected
40 apple varieties from various parts of the country are chosen as experimental raw, fructescence samples, have no mechanical damage,
No disease and pests harm；
(2) evaluation method of apple fresh food organoleptic quality grade is determined
It determines for evaluating apple fresh food organoleptic quality grade, corresponding apple is obtained according to embodiment 9 and eats sense organ product raw
Corresponding apple raw material is divided into 5 apple fresh food organoleptic quality grades by matter scoring, and the apple raw material corresponds to it and eats sense organ product raw
The grade probability of matter grade is 1, and the grade probability of remaining fresh food organoleptic quality grade is 0, specifically: 13-14.5 points are level-one,
11-12.5 is second level, and 9-10.5 is three-level, and 7-8.5 is level Four, and 4.5-6.5 is Pyatyi, is specifically shown in Table 42；
42 apple of table eats organoleptic quality grade raw
(3) apple fresh fruit Instrument measuring index is measured；Fresh fruit Instrument measuring is determined according to fresh food organoleptic quality evaluations index
Index is volume, pericarp L value, pericarp a value, pericarp b value, flesh firmness, and titratable acid, soluble sugar, crushing juice rate measures each apple
The corresponding Instrument measuring achievement data of fruit fresh fruit；
(4) neural-network learning model is constructed
35 apple samples of random screening establish learning model, residue 5 as training sample from 40 apple samples
A apple sample as forecast sample carry out apple fresh food organoleptic quality number it was predicted that mode input layer be volume, pericarp L value,
Pericarp a value, pericarp b value, flesh firmness, titratable acid, soluble sugar, 8 raw material Instrument measuring achievement datas of crushing juice rate, model
Output layer is that apple eats organoleptic quality raw, and the optimal implicit number of plies of model is 6, and maximum cycle 3000, learning rate 0.2 moves
Measure the factor 0.3, error amount 0.01；
(5) forecast sample is verified
Using neural-network learning model, the apple fresh food quality grade of 5 apple samples is predicted, as a result such as table
Shown in 41,5 apple sample predictions are accurate, predictablity rate 100%.
43 neural network prediction result of table
Apple is carried out multi-direction (dried product, fruit juice, fresh food) product as raw material and processed by above example, but the application
Raw material be not limited to apple, according to the multi-field verification experimental verification of applicant, can be also used for other fruits and vegetables, meat, agricultural production category
Processing, such as fruits include peach, banana etc., greengrocery include carrot, tomato etc., and meat products includes pork, chicken etc.,
Other agricultural product include that milk, egg etc. are not repeated them here since specification length is limited.
Number of devices and treatment scale described herein are for simplifying explanation of the invention.To application of the invention,
Modifications and variations will be readily apparent to persons skilled in the art.
Although the embodiments of the present invention have been disclosed as above, but its is not only in the description and the implementation listed
With it can be fully applied to various fields suitable for the present invention, for those skilled in the art, can be easily
Realize other modification, therefore without departing from the general concept defined in the claims and the equivalent scope, the present invention is simultaneously unlimited
In specific details and example shown and described herein.
Claims (10)
1. the method based on BP neural network prediction Raw material processing suitability characterized by comprising
The single index for evaluating the fabricated product of Raw material processing suitability is selected, is determined according to the single index of fabricated product
All multi objectives of relevant raw material；
It is quantitative progress Raw material processing by variable, processing technology and parameter of all multi objectives of raw material, records and arrange history number
According to formed sample, the format of sample are as follows: each historical data include raw material all multiple parameter datas and fabricated product it is single
Achievement data；
A certain proportion of historical data is selected to form training sample set, remaining historical data forms detection sample set；
Using all multiple parameter datas of the raw material of training sample set as input layer, with the single index of the fabricated product of training sample set
Data are output layer, determine input layer number, hidden layer neuron number and output layer neuron number, determine training
Number, learning rate, momentum, error amount determine the activation primitive of hidden layer, output layer, construct the artificial neuron based on BP algorithm
Network model, and be trained to artificial nerve network model and stablize；
It is tested with detecting sample set to artificial nerve network model, optimized artificial neural network model；
Using all multi objectives of raw material as input layer, the single finger of fabricated product is predicted according to the artificial nerve network model of optimization
Mark.
2. the method as described in claim 1 based on BP neural network prediction Raw material processing suitability, which is characterized in that
The historical data of 70-90% is selected to form training sample set, remaining data form detection sample set.
3. the method as described in claim 1 based on BP neural network prediction Raw material processing suitability, which is characterized in that
Sample includes at least 30 historical datas.
4. the method as described in claim 1 based on BP neural network prediction Raw material processing suitability, which is characterized in that
Frequency of training is 500-8000 times, learning rate 0.1-0.5, momentum 0.1-0.8, error amount 0.00001-0.1.
5. the method as described in claim 1 based on BP neural network prediction Raw material processing suitability, which is characterized in that
The index of raw material includes base values, physical and chemical index and processing index, and base values includes the volume, density, color of raw material, reason
Change index include the sugar of raw material, acid, protein, phenols, pectin content, polyphenol oxidase activity, crude fibre, crude fat, potassium,
Calcium, magnesium, processing index include edible ratio, fruit shape index, the brown stain degree of raw material.
6. the method as described in claim 1 based on BP neural network prediction Raw material processing suitability, which is characterized in that
The index of fabricated product is that the turbidity of juice, stability, comprehensive score, quality grade be one of or the color of dried product
Poor a value, brittleness, comprehensive score, quality grade or the comprehensive score of fresh food, quality grade are one of.
7. the method as described in claim 1 based on BP neural network prediction Raw material processing suitability, which is characterized in that
When determining all multi objectives of relevant raw material according to the single index of fabricated product, artificial screening goes out feedstock specifications first, then
Correlation analysis is carried out to the single index of each single item feedstock specifications and fabricated product, removal related coefficient is lower than preset correlation
The feedstock specifications of coefficient threshold.
8. the method as described in claim 1 based on BP neural network prediction Raw material processing suitability, which is characterized in that
Being lower than preset error threshold with training sample set training artificial nerve network model to error is artificial nerve network model
Stablize.
9. the method as described in claim 1 based on BP neural network prediction Raw material processing suitability, which is characterized in that
When detecting sample set test artificial nerve network model, compares the predicted value and measured value of the single index of fabricated product, accord with
Preset threshold value is closed, then artificial nerve network model is stablized, otherwise this historical data is added to training sample set and is instructed
Practice, optimized artificial neural network model.
10. the method as claimed in claim 9 based on BP neural network prediction Raw material processing suitability, feature exist
In, with detect sample set test artificial nerve network model to fabricated product single index predicted value and measured value error
Being higher than preset correlation coefficient threshold lower than preset error threshold or related coefficient is that artificial nerve network model is stablized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811147405.2A CN109409579A (en) | 2018-09-29 | 2018-09-29 | The method of BP neural network prediction Raw material processing suitability |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811147405.2A CN109409579A (en) | 2018-09-29 | 2018-09-29 | The method of BP neural network prediction Raw material processing suitability |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109409579A true CN109409579A (en) | 2019-03-01 |
Family
ID=65466575
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811147405.2A Pending CN109409579A (en) | 2018-09-29 | 2018-09-29 | The method of BP neural network prediction Raw material processing suitability |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109409579A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111325404A (en) * | 2020-02-25 | 2020-06-23 | 武汉轻工大学 | Data-driven model analysis method and device for solving rice flour raw material index range |
| CN111340361A (en) * | 2020-02-25 | 2020-06-26 | 武汉轻工大学 | Data-driven model analysis method and device for solving index range of yellow rice wine raw materials |
| CN111340369A (en) * | 2020-02-25 | 2020-06-26 | 武汉轻工大学 | Data-driven model analysis method and device for solving index range of food raw materials |
| CN111695743A (en) * | 2020-06-23 | 2020-09-22 | 信阳农林学院 | Industrial food safety production monitoring method, device, equipment and storage medium |
| CN117229905A (en) * | 2023-11-15 | 2023-12-15 | 山东朝辉生物科技有限公司 | Biological feed fermentation control method and system |
| CN117668663A (en) * | 2023-12-12 | 2024-03-08 | 广东省农业科学院蔬菜研究所 | Carrot fleshy root brittleness determination and grade prediction method based on texture analyzer and artificial neural network |
| CN117935995A (en) * | 2024-03-21 | 2024-04-26 | 江苏众钠能源科技有限公司 | Hard carbon material screening method and device for ion battery |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007106942A1 (en) * | 2006-03-21 | 2007-09-27 | Commonwealth Scientific And Industrial Research Organisation | Analysis of grape quality using neural network |
| CN103954734A (en) * | 2014-04-08 | 2014-07-30 | 广东省农业科学院蚕业与农产品加工研究所 | Quality determination and evaluation method of lichee suitable for fruit juice processing |
| CN108009639A (en) * | 2017-12-13 | 2018-05-08 | 重庆邮电大学 | A kind of city ecology construction evaluation method based on GA-BP neural network algorithms |
-
2018
- 2018-09-29 CN CN201811147405.2A patent/CN109409579A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007106942A1 (en) * | 2006-03-21 | 2007-09-27 | Commonwealth Scientific And Industrial Research Organisation | Analysis of grape quality using neural network |
| CN103954734A (en) * | 2014-04-08 | 2014-07-30 | 广东省农业科学院蚕业与农产品加工研究所 | Quality determination and evaluation method of lichee suitable for fruit juice processing |
| CN108009639A (en) * | 2017-12-13 | 2018-05-08 | 重庆邮电大学 | A kind of city ecology construction evaluation method based on GA-BP neural network algorithms |
Non-Patent Citations (2)
| Title |
|---|
| 聂琼: "基于BP人工神经网络的纱线强度CV值预测研究", 《轻纺工业与技术》 * |
| 聂继云等: "苹果品种用于加工鲜榨汁的适宜性评价", 《农业工程学报》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340361B (en) * | 2020-02-25 | 2023-04-28 | 武汉轻工大学 | Data-driven model analysis method and device for solving yellow wine raw material index range |
| CN111340361A (en) * | 2020-02-25 | 2020-06-26 | 武汉轻工大学 | Data-driven model analysis method and device for solving index range of yellow rice wine raw materials |
| CN111340369A (en) * | 2020-02-25 | 2020-06-26 | 武汉轻工大学 | Data-driven model analysis method and device for solving index range of food raw materials |
| CN111325404A (en) * | 2020-02-25 | 2020-06-23 | 武汉轻工大学 | Data-driven model analysis method and device for solving rice flour raw material index range |
| CN111340369B (en) * | 2020-02-25 | 2021-02-02 | 武汉轻工大学 | Data-driven model analysis method and device for solving index range of food raw materials |
| CN111325404B (en) * | 2020-02-25 | 2022-04-22 | 武汉轻工大学 | Data-driven model analysis method and device for solving rice flour raw material index range |
| CN111695743A (en) * | 2020-06-23 | 2020-09-22 | 信阳农林学院 | Industrial food safety production monitoring method, device, equipment and storage medium |
| CN111695743B (en) * | 2020-06-23 | 2024-04-05 | 信阳农林学院 | Industrial food safety production monitoring method, device, equipment and storage medium |
| CN117229905A (en) * | 2023-11-15 | 2023-12-15 | 山东朝辉生物科技有限公司 | Biological feed fermentation control method and system |
| CN117229905B (en) * | 2023-11-15 | 2024-02-06 | 山东朝辉生物科技有限公司 | Biological feed fermentation control method and system |
| CN117668663A (en) * | 2023-12-12 | 2024-03-08 | 广东省农业科学院蔬菜研究所 | Carrot fleshy root brittleness determination and grade prediction method based on texture analyzer and artificial neural network |
| CN117935995A (en) * | 2024-03-21 | 2024-04-26 | 江苏众钠能源科技有限公司 | Hard carbon material screening method and device for ion battery |
| CN117935995B (en) * | 2024-03-21 | 2024-06-11 | 江苏众钠能源科技有限公司 | Hard carbon material screening method and device for ion battery |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109409579A (en) | The method of BP neural network prediction Raw material processing suitability | |
| Pereira et al. | Predicting the ripening of papaya fruit with digital imaging and random forests | |
| Batu | Determination of acceptable firmness and colour values of tomatoes | |
| Nyasordzi et al. | Utilizing the IAD index to determine internal quality attributes of apples at harvest and after storage | |
| Bavay et al. | Importance of fruit variability in the assessment of apple quality by sensory evaluation | |
| CN109242341A (en) | Method based on apple feedstock specifications prediction fruit juice integrated quality | |
| CN109325626A (en) | Method based on apple feedstock specifications prediction dried product integrated quality | |
| CN103954734B (en) | A kind of mensuration of lichee quality of suitable fruit juice production and evaluation method | |
| Soares et al. | Distance-dependent competition measures for eucalyptus plantations in Portugal | |
| Giongo et al. | Raspberry texture mechanical profiling during fruit ripening and storage | |
| Gatti et al. | Integrating sensory analysis and hedonic evaluation for apple quality assessment | |
| CN112163779A (en) | Method for evaluating quality of pear fruits | |
| Zhang et al. | Changes in the chlorophyll absorbance index (I AD) are related to peach fruit maturity | |
| Ciacciulli et al. | Identification of a melting type variant among peach (P. persica L. Batsch) fruit textures by a digital penetrometer | |
| Lanza et al. | Panel performance, discrimination power of descriptors, and sensory characterization of table olive samples | |
| KR100800597B1 (en) | Method for measuring sugar content of fuji apple and it's measuring board | |
| Le Moigne et al. | How to follow grape maturity for wine professionals with a seasonal judge training? | |
| Zhou et al. | Consumer‐assisted selection: the preference for new tablegrape cultivars in C hina | |
| CN113780845A (en) | Sweet potato germplasm resource evaluation method based on multi-criterion decision | |
| CN112967807B (en) | System, device and storage medium for predicting cerebral apoplexy based on diet behavior | |
| Nantongo et al. | End‐user preferences to enhance prospects for varietal acceptance and adoption in potato breeding in Uganda | |
| Raihen et al. | Prediction modeling using deep learning for the classification of grape-type dried fruits | |
| CN112666325A (en) | Construction method and application of sensory evaluation model for chicken quality | |
| Letaief et al. | Sensory and instrumental texture measurements for assessing grape seed parameters during fruit development | |
| Naryal et al. | Sensory attributes and consumer appreciation of fresh apricots with white seed coats |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |