CN107704883A - A kind of sorting technique and system of the grade of magnesite ore - Google Patents
A kind of sorting technique and system of the grade of magnesite ore Download PDFInfo
- Publication number
- CN107704883A CN107704883A CN201710954112.4A CN201710954112A CN107704883A CN 107704883 A CN107704883 A CN 107704883A CN 201710954112 A CN201710954112 A CN 201710954112A CN 107704883 A CN107704883 A CN 107704883A
- Authority
- CN
- China
- Prior art keywords
- learning machine
- extreme learning
- elm
- grade
- data matrix
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Sorting Of Articles (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention provides a kind of sorting technique and system of the grade of magnesite ore, for providing a kind of mode of the grade of quantitative analysis magnesite ore.This method includes：Dimension-reduction treatment is carried out by stack autoencoder network to the first data matrix of magnesite ore to be sorted, obtains the second data matrix；Wherein, the first data matrix is used for the spectroscopic data for indicating the magnesite ore to be sorted；The classification of data is carried out by extreme learning machine ELM to second data matrix, to obtain the grade of the magnesite ore to be sorted；Wherein, obtained after the advance trained sample trainings of the extreme learning machine ELM, using the second data matrix as input, the model using the grade of magnesite ore as output.The method of the present invention has certain advantage in economy, accuracy and rapidity, also, this method can realize the high-volume on-line checking of ore.
Description
Technical field
The present invention relates to ore detection technique field, more particularly to a kind of sorting technique of grade of magnesite ore and it is
System.
Background technology
Magnesite ore is mainly used in the industries such as metallurgy, chemical industry, and its purity directly affects product quality.Magnesite ore
Purity can also be weighed with the grade of magnesite ore, and grade is higher, and purity is higher.At present to the side of magnesite ore classification
Method generally use chemical method is differentiated that this method needs professional staff to operate, and analytical cycle is grown, working strength
Greatly.
The content of the invention
The embodiment of the present invention provides a kind of sorting technique and system of the grade of magnesite ore, a kind of quantitative for providing
Analyze the mode of the grade of magnesite ore.
First aspect, there is provided a kind of sorting technique of the grade of magnesite ore, including：
Dimension-reduction treatment is carried out by stack autoencoder network to the first data matrix of magnesite ore to be sorted, obtained
Second data matrix；Wherein, the first data matrix is used for the spectroscopic data for indicating the magnesite ore to be sorted；
The classification of data is carried out by extreme learning machine ELM to second data matrix, it is described to be sorted to obtain
The grade of magnesite ore；Wherein, obtained after the advance trained sample trainings of the extreme learning machine ELM, with the second number
It is input according to matrix, the model using the grade of magnesite ore as output.
Optionally, the excitation function between the input layer and hidden layer of the extreme learning machine ELM is Sigmoid functions.
Optionally, the number of nodes of the input layer of the extreme learning machine ELM and hidden layer is 45.
Optionally, the building process of the extreme learning machine ELM includes：
Cycle limit learning machine ELM；Wherein, extreme learning machine ELM cycle-index is no more than N, and N is positive integer；
Compare output result of the training sample after extreme learning machine ELM training；
The one group or several groups input weights and hidden layer between input layer and hidden layer corresponding to being determined from output result
Threshold value；
The model being made up of the input weights determined and the hidden layer threshold value is redefined as the limit
Habit machine ELM.
Optionally, N is less than or equal to 200.
Optionally, the classification of data is carried out by extreme learning machine ELM to second data matrix, to obtain described treat
The grade of the magnesite ore of classification, including：
Second data matrix is inputted into the extreme learning machine ELM, through the excitation function and the input
The target output matrix of the hidden layer is obtained after weights computing；
Using the target output matrix and grade of the product as the magnesite ore of output for exporting weight.
Second aspect, there is provided a kind of categorizing system of the grade of magnesite ore, including：
Dimensionality reduction unit, for being carried out to the first data matrix of magnesite ore to be sorted by stack autoencoder network
Dimension-reduction treatment, obtain the second data matrix；Wherein, the first data matrix is used for the light for indicating the magnesite ore to be sorted
Modal data；
Output unit, for carrying out the classification of data by extreme learning machine ELM to second data matrix, to obtain
The grade of the magnesite ore to be sorted；Wherein, the extreme learning machine ELM is to obtain after advance trained sample training
, using the second data matrix as input, the model using the grade of magnesite ore as output.
Optionally, the excitation function between the input layer and hidden layer of the extreme learning machine ELM is Sigmoid functions.
Optionally, the number of nodes of the input layer of the extreme learning machine ELM and hidden layer is 45.
Optionally, the system also includes：
Model unit is built, for cycle limit learning machine ELM；Wherein, extreme learning machine ELM cycle-index is no more than
N, N are positive integer；Compare output result of the training sample after extreme learning machine ELM training；Determined from output result
The corresponding one group or several groups input weights and hidden layer threshold value between input layer and hidden layer；It will be weighed by the input determined
The model of value and hidden layer threshold value composition is redefined as the extreme learning machine ELM.
The sorting technique of the grade of magnesite ore provided in an embodiment of the present invention, by fisrt feature parameter to be identified with
Multiple characteristic parameters of storage are compared, and obtain multiple similarities, multiple characteristic parameters are may determine that according to multiple similarities
Corresponding objective attribute target attribute, identifies which destination object multiple characteristic parameters belong to.If maximum similarity is less than default
First threshold, then be possible to not identify or identify inaccuracy, can continue further to be identified, improve identification essence
Degree.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the required accompanying drawing used in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Inventive embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, can also be according to carrying
The accompanying drawing of confession obtains other accompanying drawings.
Fig. 1 is the flow chart of the classification of the grade of magnesite ore provided in an embodiment of the present invention；
Fig. 2 is a kind of structural representation of autoencoder network provided in an embodiment of the present invention；
Fig. 3 is ELM training sets output result provided in an embodiment of the present invention and actual desired scatter diagram；
Fig. 4 is ELM test sets output result provided in an embodiment of the present invention and actual desired scatter diagram；
Fig. 5 is selected ELM training sets output result provided in an embodiment of the present invention and actual desired scatter diagram；
Fig. 6 is selected ELM test sets output result provided in an embodiment of the present invention and actual desired scatter diagram；
Fig. 7 is distributed for integrated-selected ELM training sets output result provided in an embodiment of the present invention with actual desired scatterplot
Figure；
Fig. 8 is distributed for integrated-selected ELM test sets output result provided in an embodiment of the present invention with actual desired scatterplot
Figure；
Fig. 9 is a kind of structural representation of the categorizing system of the grade of magnesite ore provided in an embodiment of the present invention.
Embodiment
For the object, technical solutions and advantages of the present invention are more clearly understood, below in conjunction with the embodiment of the present invention
Accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is only
It is part of the embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, ordinary skill people
The every other embodiment that member is obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.
In the case of not conflicting, the feature in embodiment and embodiment in the present invention can be mutually combined.Although also, flowing
Logical order is shown in journey figure, but in some cases, can be with different from shown or described by order execution herein
The step of.
In addition, the terms "and/or", only a kind of incidence relation for describing affiliated partner, represents there may be
Three kinds of relations, for example, A and/or B, can be represented：Individualism A, while A and B be present, these three situations of individualism B.Separately
Outside, character "/" herein, in the case where not illustrating, it is a kind of relation of "or" to typically represent forward-backward correlation object.
In order to more fully understand above-mentioned technical proposal, below in conjunction with Figure of description and specific embodiment to upper
Technical scheme is stated to be described in detail.
Refer to Fig. 1, there is provided a kind of sorting technique of the grade of magnesite ore, the flow of this method are described as follows.
S101：First data matrix of magnesite ore to be sorted is carried out at dimensionality reduction by stack autoencoder network
Reason, obtains the second data matrix；Wherein, the first data matrix is used for the spectrum number for indicating the magnesite ore to be sorted
According to；
S102：The classification of data is carried out by extreme learning machine ELM to second data matrix, to obtain described treat point
The grade of the magnesite ore of class；Wherein, the extreme learning machine ELM is obtained after trained sample training in advance, with the
Two data matrixes are input, the model using the grade of magnesite ore as output.
First data matrix can be used for the characteristic parameter for characterizing magnesite ore to be sorted, can pass through infrared spectrum
Instrument is tested magnesite ore to be sorted, obtains the spectroscopic data of magnesite ore.In the embodiment of the present invention, pass through light
Spectrometer gathers spectroscopic data to the magnesite ore that sample size is 531 parts, and the original spectral data gathered is a dimension
For 973 high dimensional data matrix.Because the dimension of the first data matrix is higher, and data dimension is higher, and this results in data tool
Have larger noise, and common factor there may be between data, interfere with each other, therefore, to spectroscopic data input disaggregated model it
It is preceding, it is necessary to carry out dimension-reduction treatment to data, to reduce data processing amount, improve data-handling efficiency.
In the embodiment of the present invention, dimension-reduction treatment is carried out by stack autoencoder network to the first data matrix, after processing
To data matrix be referred to as the second data matrix.Because the data included by the first data matrix there may be nonlinear organization,
So conventional Method of Data with Adding Windows, the method for the suitable linear data dimensionality reduction of such as PCA, ICA are obviously not suitable for.Therefore, it is of the invention
Embodiment carries out dimension-reduction treatment using autoencoder network (SAE) to the first data matrix, specifically can be defeated by construction successively
Enter layer to carry out dimensionality reduction and feature extraction to data equal to the three-layer neural network of output layer.Fig. 2 is referred to, Fig. 2 is own coding
The structural representation of network.
In fig. 2, self-encoding encoder is attempted to go to learn an identity function, and the identity function illustrates with formula (1)
hw,b(x)≈x (1)
Two steps of coding and decoding can be divided into the pre-training stage of autoencoder network.First layer belongs to the second layer
Cataloged procedure, it is exactly to define a Function Mapping g () in the process so that input data x, which is converted into hide, represents u, such as
Shown in formula (2).
U=g (Wx+b0) (2)
Wherein, W, b are weights and deviation of the input layer to hidden layer respectively, and W is the weight matrix of a d × d ' dimensions.b
It is d ' dimension bias vectors.Function g () is a Nonlinear Mapping.
The second layer belongs to decoding process to third layer, and u is reconstructed by g () to obtain hw,b(x), such as formula (3)：
hw,b(x)=g (WTu+b1) (3)
Optimize parameter W, b of own coding model by minimizing average reconstruction error0And b1.Average reconstruction error is determined
Adopted form is various, in the embodiment of the present invention, from formula (4).
L(x,hW,b(x))=| | x-hW,b(x)||2 (4)
In the embodiment of the present invention, renewal is iterated to parameter W and b using gradient descent method and back-propagation algorithm, from
And learn the depth autoencoder network optimized.
For fixed sample collection { (x(1),y(1)),····,(x(m),y(m)), it contains m sample.When setting hidden layer section
When the s that counts is smaller than input layer number m, we can obtain the compression expression of an input.We can define entirety
Cost function such as formula (5).
Section 1 J (W, b) in formula (5) is a mean square deviation item.Section 2 is a regularization term.λ declines for weight
Subtract parameter, the purpose is to reduce the amplitude of weight, prevent overfitting.We use nlTo represent the number of plies of network.Wherein, Wij (l)
It is the parameter that couples between l layer jth units and l+1 layer i-th cells, bi (l)It is the bias term of l+1 layer i-th cells.When
When setting node in hidden layer s is bigger than input layer number m, by adding some openness restrictive conditions, obtain sparse
The result of coding.Now total cost function can use formula (6) to represent.
Wherein,Be one using ρ as average and one withFor average
Two Bernoulli random variables between relative entropy.
Iteration is all updated according to formula (7) to parameter W and b each time in gradient descent method.
Wherein,Wherein, α is learning rate.
The embodiment of the present invention, use stack autoencoder network (SAE) to pretreated first data matrix (dimension for
973) dimension-reduction treatment is carried out.Wherein for hidden layer altogether by having two layers, the first hidden layer has 200 node numbers, and the second hidden layer has
100 node numbers, that is, by dimension-reduction treatment by original giobertite near infrared spectrum data dimensionality reduction to 100.
After the spectroscopic data progress dimension-reduction treatment of magnesite ore, extreme learning machine ELM can be passed through and carry out data
Classification, to obtain the grade of the magnesite ore to be sorted；The extreme learning machine ELM is advance trained sample training
Obtain afterwards, using the second data matrix as input, the model using the grade of magnesite ore as output.
ELM is the algorithm that traditional neural network parameter optimization is solved to system of linear equations replacement iteration.With traditional
Practise algorithm to compare, ELM overcomes the successive ignition of parameter, therefore ELM has faster pace of learning and good generalization ability.
For any given N number of different sample (xi,ti), xi=[xi1,xi2,…,xin]T∈Rn, ti=[ti1,
ti2,…,tim]T∈Rm, then represent that the Single hidden layer feedforward neural networks such as formula (8) of L hidden node is shown.
Wherein, x ∈ R, ai∈Rn,βi∈Rm, G (ai,bi, x) and represent i-th of hidden node and x relation.g(x)：R → R,
Draw
G(ai,bi, x) and=g (ai·x+bi),bi∈R (9)
Wherein, ai=[ai1,ai2,…,ain]T∈RnInput weight vector for input layer to i-th of hidden layer node, bi
Represent the threshold value of i-th of hidden layer node, βi=[βi1,βi2,…βim]TRepresent output of i-th of hidden layer node to output layer
Weight vector.ai, biIt is random value in the training of model.
Choose N number of sample (xi,ti)∈Rn×Rm, xi∈Rn, ti∈Rm.Formula (8) can be simplified to：
H β=T (10)
Formula (10), H are hidden layer output matrixes.
Finally, the minimum value of the least-square solution of linear system is：
Formula (11),It is H generalized inverse expression-form.
Extreme learning machine ELM models are described below establishes process.
The experiment sample spectroscopic data treated through SAE is 531 × 100, the instruction of random-selection model in 531 groups of data
Practice collection and test set.Training sample totally 367, wherein comprising select quality 72, primes 73, seconds 72, three-level product 80 and giving up
Ore deposit 70.Totally 164, test set sample, wherein select quality 32, primes 36, seconds 36, three-level product 30 and abandoned mine 30.ELM water chestnuts
The parameter selection of magnesium ore grade disaggregated model includes activation primitive and node in hidden layer.And ELM activation primitive generally has
It is several below：Sigmoid functions, sin functions, hardlim functions etc..The embodiment of the present invention selects Sigmoid functions as ELM
Giobertite grade disaggregated model in activation primitive, obtained structure is more accurate.Because node in hidden layer can be to net
The study of network and information processing capability produce large effect.The nodes complexity on the high side that can increase network so that study
Time is elongated, and over-fitting easily occurs.Nodes study that then can be to the information of network on the low side and disposal ability produce one
Fixed restriction.In the embodiment of the present invention, node in hidden layer can be located at first in the range of, the first scope cause learning time compared with
It is short, and the over-fitting imagination will not occur, the first scope can determine a probable ranges, hidden layer node using empirical equation
Number can be an optimal value in the range of first, if node in hidden layer is 45.Calculated by formula (11) and obtain output weights,
And unique optimal solution can be obtained.The training set and test set output result of ELM algorithm models and actual desired scatterplot point
Butut difference is as shown in Figure 3, Figure 4.
Traditional ELM algorithm models are for giobertite grade classifying quality unobvious it can be seen from Fig. 3 and Fig. 4.Due to
ELM inputs weights and hidden layer threshold value random assignment, and ELM output is unstable, and being easily trapped into local minimum causes accuracy rate
It is not high.
In consideration of it, the embodiment of the present invention proposes a kind of ELM innovatory algorithm, selected ELM is properly termed as.By extreme learning machine
ELM circulates n times, and N is positive integer, and N selection can be with experience or experiment gained so that extreme learning machine ELM output results are more
To be accurate.By the output result of comparing cell training set, find one group or several groups input weights of accuracy rate highest and imply
Layer threshold value is simultaneously remained.It is fixed as ELM model parameters, that is, obtains selected ELM giobertites disaggregated model.Essence
Select ELM models training set and test set output result with actual desired scatter diagram as shown in Figure 5, Figure 6.
Selected ELM prediction result has obvious this to enter than traditional ELM as can be known from Fig. 5 and Fig. 6.Each group of essence
The corresponding one group of parameter of ELM is selected, the result that each selected ELM model differentiates for same sample might have difference, and its is steady
It is qualitative not high.In order to improve the stability of modelling effect, this paper presents integrated-selected ELM algorithms.Choose herein integrated 11 groups
Selected model, the output of integrated model is that 11 single models export most categories.This can further improve the prediction of model
Precision.The quantity between the predicted value of grade and desired value is corresponded to by the giobertite in experiment simulation and statistical simulation result
Relation, it can be seen that the stability and accuracy that integrated-selected ELM models are classified in giobertite grade obtain good effect
Fruit.The training set of integrated-selected ELM models and the statistics of the contrast of test set Different categories of samples emulation accurate result and actual sample
Figure is as shown in Figure 7, Figure 8, more accurate compared to selected ELM models output result.
The embodiment of the present invention, for giobertite grade data, we establish common ELM respectively, essence takes ELM and collection
ELM disaggregated models are taken into-essence.Its simulation result is as shown in table 1 below.
Table 1ELM, essence take ELM and integrated-essence to take ELM classification simulation results
| Types of models | Training set accuracy | Test set accuracy | Model takes |
| ELM (%) | 86.921% | 78.0488% | 0.2257s |
| Selected ELM (%) | 96.1853% | 87.8049% | 1.22s |
| Integrated-selected ELM (%) | 99.7275% | 98.1707% | 48.2s |
From table 1 it follows that the ELM models of giobertite grade classification, essence take ELM models, integrated-selected ELM moulds
The degree of accuracy of type.Conventional ELM model measurement collection accuracys rate are not high, and model is time-consuming few.Selected ELM precision is significantly improved,
Its accuracy can generally achieve more than 85%, but its stability is not high.Integrated-selected ELM is steady simulation accuracy and model
Qualitative aspect tool improves a lot, and precision can reach 98%.
The embodiment of the present invention proposes a kind of sorting technique of the grade of new magnesite ore.It is primarily based near infrared light
The non-destructive testing technology of spectrum, gather the spectroscopic data of ore.Then dimensionality reduction is carried out using SAE methods.Finally calculated using selected ELM
Method is modeled, and accuracy rate is higher.For selected ELM prediction stability it is bad the problem of, the embodiment of the present invention further carries
Go out integrated-selected ELM methods.No matter integrated-selected ELM is obvious in the accuracy or stability that giobertite grade is classified
Better than traditional ELM and selected ELM, predictablity rate is up to 98%.Compared with traditional giobertite grade sorting technique, herein
The giobertite detection method of proposition has certain advantage in economy, accuracy and rapidity.Also, this method
The high-volume on-line checking of ore can be realized.It can be seen that it has important actual application value.
The system that the embodiment of the present invention is provided is introduced below in conjunction with the accompanying drawings.
Fig. 9 is referred to, based on same inventive concept, the embodiment of the present invention also provides a kind of point of grade of magnesite ore
Class system, the categorizing system include dimensionality reduction unit 901, output unit 902 and structure model unit 903.
Dimensionality reduction unit 901 can be used for passing through stack own coding net to the first data matrix of magnesite ore to be sorted
Network carries out dimension-reduction treatment, obtains the second data matrix；Wherein, the first data matrix is used to indicate the magnesite ore deposit to be sorted
The spectroscopic data of stone；
Output unit 902 can be used for the classification for carrying out data by extreme learning machine ELM to second data matrix,
To obtain the grade of the magnesite ore to be sorted；Wherein, the extreme learning machine ELM is advance trained sample training
Obtain afterwards, using the second data matrix as input, the model using the grade of magnesite ore as output.
Optionally, the excitation function between the input layer and hidden layer of the extreme learning machine ELM is Sigmoid functions.
Optionally, the number of nodes of the input layer of the extreme learning machine ELM and hidden layer is 45.
Optionally, structure model unit 903 can be used for cycle limit learning machine ELM；Wherein, extreme learning machine ELM
Cycle-index is no more than N, and N is positive integer；Compare output result of the training sample after extreme learning machine ELM training；From
The one group or several groups input weights and hidden layer threshold value between input layer and hidden layer corresponding to being determined in output result；Will be by true
The model of the fixed input weights and hidden layer threshold value composition is redefined as the extreme learning machine ELM.
The categorizing system of the grade of the magnesite ore can be used for performing the magnesite ore deposit described in above-mentioned Fig. 1 embodiments
The sorting technique of the grade of stone, therefore, function of being realized for each module in the categorizing system etc., refer to such as foregoing side
The description of method part, is seldom repeated.
It is apparent to those skilled in the art that for convenience and simplicity of description, only with above-mentioned each function
The division progress of module, can be as needed and by above-mentioned function distribution by different function moulds for example, in practical application
Block is completed, i.e., the internal structure of device is divided into different functional modules, to complete all or part of work(described above
Energy.The specific work process of the system, apparatus, and unit of foregoing description, it may be referred to corresponding in preceding method embodiment
Journey, it will not be repeated here.
In several embodiments provided by the present invention, it should be understood that disclosed apparatus and method, it can be passed through
Its mode is realized.For example, device embodiment described above is only schematical, for example, the module or unit
Division, only a kind of division of logic function, can there is other dividing mode, such as multiple units or component when actually realizing
Another system can be combined or be desirably integrated into, or some features can be ignored, or do not perform.It is another, it is shown or
The mutual coupling discussed or direct-coupling or communication connection can be the indirect couplings by some interfaces, device or unit
Close or communicate to connect, can be electrical, mechanical or other forms.
The unit illustrated as separating component can be or may not be physically separate, show as unit
The part shown can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On NE.Some or all of unit therein can be selected to realize the mesh of this embodiment scheme according to the actual needs
's.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, can also
That unit is individually physically present, can also two or more units it is integrated in a unit.Above-mentioned integrated list
Member can both be realized in the form of hardware, can also be realized in the form of SFU software functional unit.
If the integrated unit is realized in the form of SFU software functional unit and is used as independent production marketing or use
When, it can be stored in a computer read/write memory medium.Based on such understanding, technical scheme is substantially
The part to be contributed in other words to prior art or all or part of the technical scheme can be in the form of software products
Embody, the computer software product is stored in a storage medium, including some instructions are causing a computer
It is each that equipment (can be personal computer, server, or network equipment etc.) or processor (processor) perform the present invention
The all or part of step of embodiment methods described.And foregoing storage medium includes：USB flash drive
(Universal Serial Bus flash drive, USB flash drive), mobile hard disk, read-only storage (Read-
Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc. it is various
Can be with the medium of store program codes.
Specifically, computer program instructions can be stored in corresponding to the target identification method in the embodiment of the present application
In the storage mediums such as CD, hard disk, USB flash disk, counted when the sorting technique of the grade with magnesite ore in storage medium is corresponding
When calculation machine programmed instruction is read or is performed by the categorizing system of the grade of a magnesite ore, comprise the following steps：
Dimension-reduction treatment is carried out by stack autoencoder network to the first data matrix of magnesite ore to be sorted, obtained
Second data matrix；Wherein, the first data matrix is used for the spectroscopic data for indicating the magnesite ore to be sorted；
The classification of data is carried out by extreme learning machine ELM to second data matrix, it is described to be sorted to obtain
The grade of magnesite ore；Wherein, obtained after the advance trained sample trainings of the extreme learning machine ELM, with the second number
It is input according to matrix, the model using the grade of magnesite ore as output.
Optionally, also it is stored with the storage medium between the input layer and hidden layer of the extreme learning machine ELM
Excitation function is Sigmoid functions.
Optionally, the input layer of the extreme learning machine ELM and the node of hidden layer are also stored with the storage medium
Quantity is 45.
Optionally, other computer instruction is also stored with the storage medium, these computer instructions are in step：
It is performed in the building process of the extreme learning machine ELM, comprises the following steps when executed：
Cycle limit learning machine ELM；Wherein, extreme learning machine ELM cycle-index is no more than N, and N is positive integer；
Compare output result of the training sample after extreme learning machine ELM training；
The one group or several groups input weights and hidden layer between input layer and hidden layer corresponding to being determined from output result
Threshold value；
The model being made up of the input weights determined and the hidden layer threshold value is redefined as the limit
Habit machine ELM.
Optionally, other computer instruction is also stored with the storage medium, these computer instructions are in step：
The classification of data is being carried out by extreme learning machine ELM to second data matrix, to obtain the magnesite to be sorted
It is performed during the grade of ore, comprises the following steps when executed：
Second data matrix is inputted into the extreme learning machine ELM, through the excitation function and the input
The target output matrix of the hidden layer is obtained after weights computing；
Using the target output matrix and grade of the product as the magnesite ore of output for exporting weight.
Obviously, those skilled in the art can carry out the essence of various changes and modification without departing from the present invention to the present invention
God and scope.So, if these modifications and variations of the present invention belong to the scope of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to comprising including these changes and modification.
Claims (10)
1. a kind of sorting technique of the grade of magnesite ore, including：
Dimension-reduction treatment is carried out by stack autoencoder network to the first data matrix of magnesite ore to be sorted, obtains second
Data matrix；Wherein, the first data matrix is used for the spectroscopic data for indicating the magnesite ore to be sorted；
The classification of data is carried out by extreme learning machine ELM to second data matrix, to obtain the magnesite to be sorted
The grade of ore deposit ore；Wherein, obtained after the advance trained sample trainings of the extreme learning machine ELM, with the second data square
Battle array is input, the model using the grade of magnesite ore as output.
2. the method as described in claim 1, it is characterised in that between the input layer and hidden layer of the extreme learning machine ELM
Excitation function be Sigmoid functions.
3. the method as described in claim 1, it is characterised in that the input layer of the extreme learning machine ELM and the section of hidden layer
Point quantity is 45.
4. method as claimed in claim 2, it is characterised in that the building process of the extreme learning machine ELM includes：
Cycle limit learning machine ELM；Wherein, extreme learning machine ELM cycle-index is no more than N, and N is positive integer；
Compare output result of the training sample after extreme learning machine ELM training；
The one group or several groups input weights and hidden layer threshold value between input layer and hidden layer corresponding to being determined from output result；
The model being made up of the input weights determined and the hidden layer threshold value is redefined as the extreme learning machine
ELM。
5. method as claimed in claim 4, it is characterised in that N is less than or equal to 200.
6. method as claimed in claim 4, it is characterised in that entered to second data matrix by extreme learning machine ELM
The classification of row data, to obtain the grade of the magnesite ore to be sorted, including：
Second data matrix is inputted into the extreme learning machine ELM, through the excitation function and the input weights
The target output matrix of the hidden layer is obtained after computing；
Using the target output matrix and grade of the product as the magnesite ore of output for exporting weight.
7. a kind of categorizing system of the grade of magnesite ore, including：
Dimensionality reduction unit, dimensionality reduction is carried out by stack autoencoder network for the first data matrix to magnesite ore to be sorted
Processing, obtains the second data matrix；Wherein, the first data matrix is used for the spectrum number for indicating the magnesite ore to be sorted
According to；
Output unit, it is described to obtain for carrying out the classification of data by extreme learning machine ELM to second data matrix
The grade of magnesite ore to be sorted；Wherein, obtained after the advance trained sample trainings of the extreme learning machine ELM,
Using the second data matrix as input, the model using the grade of magnesite ore as output.
8. system as claimed in claim 7, it is characterised in that between the input layer and hidden layer of the extreme learning machine ELM
Excitation function be Sigmoid functions.
9. system as claimed in claim 7, it is characterised in that the input layer of the extreme learning machine ELM and the section of hidden layer
Point quantity is 45.
10. the system as described in claim 7-9 is any, it is characterised in that also include：
Model unit is built, for cycle limit learning machine ELM；Wherein, extreme learning machine ELM cycle-index is no more than N, N
For positive integer；Compare output result of the training sample after extreme learning machine ELM training；Determined from output result corresponding
The one group or several groups input weights and hidden layer threshold value between input layer and hidden layer；By by the input weights that determine and
The model of the hidden layer threshold value composition is redefined as the extreme learning machine ELM.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611044847.5A CN106485289A (en) | 2016-11-21 | 2016-11-21 | A kind of sorting technique of the grade of magnesite ore and equipment |
| CN2016110448475 | 2016-11-21 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107704883A true CN107704883A (en) | 2018-02-16 |
Family
ID=58275333
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611044847.5A Pending CN106485289A (en) | 2016-11-21 | 2016-11-21 | A kind of sorting technique of the grade of magnesite ore and equipment |
| CN201710954112.4A Pending CN107704883A (en) | 2016-11-21 | 2017-10-13 | A kind of sorting technique and system of the grade of magnesite ore |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611044847.5A Pending CN106485289A (en) | 2016-11-21 | 2016-11-21 | A kind of sorting technique of the grade of magnesite ore and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN106485289A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102012A (en) * | 2018-07-30 | 2018-12-28 | 上海交通大学 | A kind of defect identification method and system of local discharge signal |
| CN111950427A (en) * | 2020-08-06 | 2020-11-17 | 中国地质科学院矿产资源研究所 | Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model |
| CN112163591A (en) * | 2020-08-21 | 2021-01-01 | 海城市后英经贸集团有限公司菱镁矿 | Magnesite ore grade classification modeling method |
| CN112485394A (en) * | 2020-11-10 | 2021-03-12 | 浙江大学 | Water quality soft measurement method based on sparse self-coding and extreme learning machine |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108734332B (en) * | 2018-03-29 | 2021-11-26 | 浙江长兴笛卡尔科技有限公司 | Method and device for predicting electrical parameters by machine learning |
| CN108983127B (en) * | 2018-05-29 | 2019-10-29 | 东北大学 | A kind of bloodstone magnetism rate detection method based on spectroscopic data |
| CN113984707B (en) * | 2021-10-19 | 2022-09-30 | 厦门兑泰新材料科技有限公司 | Tailings intelligent efficient comprehensive utilization method and system based on joint ANN |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105205491A (en) * | 2015-08-19 | 2015-12-30 | 西安电子科技大学 | Polarized SAR image classification method based on extreme learning machine |
| CN105354347A (en) * | 2015-07-15 | 2016-02-24 | 东北大学 | Magnesite ore grade classification modeling method |
| CN105718963A (en) * | 2016-03-09 | 2016-06-29 | 东南大学 | SAR image classification method based on variable-length incremental type extreme learning machine |
-
2016
- 2016-11-21 CN CN201611044847.5A patent/CN106485289A/en active Pending
-
2017
- 2017-10-13 CN CN201710954112.4A patent/CN107704883A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105354347A (en) * | 2015-07-15 | 2016-02-24 | 东北大学 | Magnesite ore grade classification modeling method |
| CN105205491A (en) * | 2015-08-19 | 2015-12-30 | 西安电子科技大学 | Polarized SAR image classification method based on extreme learning machine |
| CN105718963A (en) * | 2016-03-09 | 2016-06-29 | 东南大学 | SAR image classification method based on variable-length incremental type extreme learning machine |
Non-Patent Citations (5)
| Title |
|---|
| VINCENTP,LAROCHELLEH,LAJOIEI,BENGIO Y,MANZAGOL P A: "Stackeddenoisingautoencoders:Learningusefulrepresentationsinadeepnetworkwithalocaldenoisingcriterion", 《JOURNALOFMACHINELEARNINGRESEARCH》 * |
| 付华: "《煤矿瓦斯突出灾害辨识技术》", 31 July 2016 * |
| 刘威等: "互学习神经网络训练方法研究", 《计算机学报》 * |
| 杨宇: "基于深度学习特征的图像推荐***", 《中国优秀硕士学位论文全文数据库》 * |
| 樊重俊等: "《大数据分析与应用》", 31 January 2016 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102012A (en) * | 2018-07-30 | 2018-12-28 | 上海交通大学 | A kind of defect identification method and system of local discharge signal |
| CN111950427A (en) * | 2020-08-06 | 2020-11-17 | 中国地质科学院矿产资源研究所 | Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model |
| CN111950427B (en) * | 2020-08-06 | 2022-02-15 | 中国地质科学院矿产资源研究所 | Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model |
| CN112163591A (en) * | 2020-08-21 | 2021-01-01 | 海城市后英经贸集团有限公司菱镁矿 | Magnesite ore grade classification modeling method |
| CN112485394A (en) * | 2020-11-10 | 2021-03-12 | 浙江大学 | Water quality soft measurement method based on sparse self-coding and extreme learning machine |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106485289A (en) | 2017-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107704883A (en) | A kind of sorting technique and system of the grade of magnesite ore | |
| Tang et al. | Very large-scale data classification based on K-means clustering and multi-kernel SVM | |
| CN109461025B (en) | Electric energy substitution potential customer prediction method based on machine learning | |
| CN110472817A (en) | A kind of XGBoost of combination deep neural network integrates credit evaluation system and its method | |
| CN110956273A (en) | Credit scoring method and system integrating multiple machine learning models | |
| CN110197205A (en) | A kind of image-recognizing method of multiple features source residual error network | |
| CN112613536B (en) | Near infrared spectrum diesel fuel brand recognition method based on SMOTE and deep learning | |
| CN112215696A (en) | Personal credit evaluation and interpretation method, device, equipment and storage medium based on time sequence attribution analysis | |
| CN104966106B (en) | A kind of biological age substep Forecasting Methodology based on support vector machines | |
| CN106681305A (en) | Online fault diagnosing method for Fast RVM (relevance vector machine) sewage treatment | |
| CN114764682B (en) | Rice safety risk assessment method based on multi-machine learning algorithm fusion | |
| CN113240201A (en) | Method for predicting ship host power based on GMM-DNN hybrid model | |
| CN115730241A (en) | Construction method of cavitation noise identification model of water turbine | |
| CN113159225B (en) | Multivariable industrial process fault classification method | |
| Rofik et al. | The Optimization of Credit Scoring Model Using Stacking Ensemble Learning and Oversampling Techniques | |
| CN109164794B (en) | Multivariable industrial process Fault Classification based on inclined F value SELM | |
| Bustillo et al. | Predictive Hybridization Model integrating Modified Genetic Algorithm (MGA) and C4. 5 | |
| CN114519508A (en) | Credit risk assessment method based on time sequence deep learning and legal document information | |
| Tiruneh et al. | Feature selection for construction organizational competencies impacting performance | |
| CN113177417A (en) | Trigger word recognition method based on hybrid neural network and multi-stage attention mechanism | |
| CN112488188A (en) | Feature selection method based on deep reinforcement learning | |
| Öztürk | Use of genetic algorithm to design optimal neural network structure | |
| CN107704952A (en) | A kind of attack of terrorism Forecasting Methodology based on stochastic subspace | |
| CN116150687A (en) | Fluid pipeline leakage identification method based on multi-classification G-WLSTSVM model | |
| Chen et al. | Graph convolution network-based fault diagnosis method for the rectifier of the high-speed train |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180216 |