CN107330474A - A kind of lithium battery cascade utilization screening method - Google Patents
A kind of lithium battery cascade utilization screening method Download PDFInfo
- Publication number
- CN107330474A CN107330474A CN201710565905.7A CN201710565905A CN107330474A CN 107330474 A CN107330474 A CN 107330474A CN 201710565905 A CN201710565905 A CN 201710565905A CN 107330474 A CN107330474 A CN 107330474A
- Authority
- CN
- China
- Prior art keywords
- lithium battery
- training
- disaggregated model
- label
- images
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Secondary Cells (AREA)
Abstract
The present invention provides a kind of lithium battery cascade utilization screening method, belongs to lossless contactless screening technique and can improve screening efficiency.Methods described includes:Obtain training sample data and training sample label;Wherein, the training sample data are the CT images of the sample lithium battery for training, and the training sample label includes:Whether the corresponding label information of CT images, the label information is used to identify the corresponding sample lithium battery for being used to train being capable of echelon utilization;It is trained according to the disaggregated model that the training sample data and training sample label of acquisition treat training, the disaggregated model after being trained;The CT images of lithium battery to be measured are obtained, the CT images of the lithium battery to be measured of acquisition are input in the disaggregated model after training, whether the disaggregated model after training exports the lithium battery to be measured being capable of echelon utilization.The present invention relates to lithium battery recycle utilization technology field.
Description
Technical field
The present invention relates to lithium battery recycle utilization technology field, a kind of lithium battery cascade utilization screening method is particularly related to.
Background technology
In recent years, China's electric automobile industry development speed, it is following in two or three years, i.e., can welcome dynamic lithium battery retired
Peak, and these lithium battery total amounts every year can also in acceleration increase;To the year two thousand twenty, electric car market volume is more than 500
Ten thousand, averagely it is equipped with 20kWh lithium battery to estimate with a car, the lithium ion lithium battery that there are about 100,000,000 kWh (1000GWh) enters
Enter automobile market.It is well known that the chemical substance and heavy metal element in lithium ion lithium battery can be polluted and endangered to environment
Evil.The surplus value that these lithium batteries are utilized with echelon how is reclaimed to greatest extent, is urgently to be resolved hurrily, very important science
Technical problem.
For retired dynamic lithium battery, there are two kinds of feasible processing methods, one kind is, directly as industrial wastes, to carry out
Scrap and disassemble, refine raw material therein, realize recycling for raw material, the enterprise that this respect there are some domestic exists
Carry out commercial operations;Another way, then consider retired dynamic lithium battery, although has been unsatisfactory for the use bar of automobile
Part, but still possess certain complementary energy, its life-span does not terminate completely, and can be used in other field makes as the carrier of electric energy
With so as to give full play to its surplus value.The echelon of obvious the latter realizes circulation using can more play the maximum value of product
Economic benefit, is more green and environmental protection approach.
Chinese invention patent CN201310261893.0 proposes a kind of waste and old dynamic lithium battery cascade utilization screening method:
(1) waste and old dynamic lithium battery group is charged, it is 15%~80% to make its state-of-charge SOC;Then lithium battery group is taken apart, it is right
Dynamic lithium battery group and single lithium battery outward appearance are checked, and are recorded;(2) open-circuit voltage of each single lithium battery is detected
And internal resistance and record, open-circuit voltage, internal resistance contrast with standard single lithium battery;By test voltage, internal resistance is according to standard list
Body charging and discharging lithium battery curve, assesses waste and old dynamic lithium battery group single lithium battery capacity;(3) by above-mentioned waste and old dynamic lithium battery
Monomer is in parallel, until its open-circuit voltage is essentially identical, with it is in parallel before the open-circuit voltage of single lithium battery contrast, and recording voltage liter
Drop situation;Then waste and old single lithium battery and standard single lithium battery are shelved 3~7 days under the conditions of temperature is 30 DEG C~55 DEG C
Or shelve at room temperature 10~30 days, its open-circuit voltage and internal resistance are detected, and record;Standard single lithium battery does discharge cycles survey
Examination, using standard single lithium battery state-of-charge, capacitance-voltage curves, internal resistance as reference, according to waste and old dynamic lithium battery monomer
Open-circuit voltage and internal resistance size, assess waste and old dynamic lithium battery monomer health status;(4) according to above record case, contrast is useless
Old dynamic lithium battery monomer show, open-circuit voltage, internal resistance, voltage drop and health state evaluation, to waste and old dynamic lithium battery monomer
It is classified, the lithium battery with one-level is used cooperatively with energy storage power network in groups.
It is related to a kind of secondary screening used of waste and old dynamic lithium battery in the A of Patent No. CN 103901350 invention
Method:First pass through and discharge and recharge is integrally carried out to lithium battery pack, the data that BMS is recorded during by testing in discharge process, are chosen
2-4 SOC point, SOC numerical value reads the voltage of each series connection monomer between 20%-90%;It will deviate from out most of lithium battery
Lithium battery of the magnitude of voltage more than 5% regard as problematic lithium battery, remaining lithium battery is tentatively to regard as health status lithium
Battery;The internal resistance of the remaining every lithium battery of measurement, carries out programmed screening by internal resistance value, will deviate from most of battery core normally interior
The battery core of resistance 20% is rejected, and screening terminates;In remaining lithium battery, 10 lithium batteries are picked out at random and carry out a discharge and recharge,
By the average size of 10 lithium batteries, the capability value of the lithium battery of all health status is defaulted as.
The screening technique used in the prior art, is all more or less related to the lithium battery parameter i.e. survey of contact
Amount, such as the detection of open-circuit voltage and internal resistance is, it is necessary to there is the process of discharge and recharge to battery;On the one hand the measurement of these parameters may
Lithium battery interior structure can be caused to change and damaged, another aspect lithium cell charging and discharge process even more need several
Hour could complete, and cause the time of consuming, cost of labor all higher, cause the recovery for being commercialized lithium battery to be utilized with echelon
Economic benefit be restricted.
The content of the invention
The technical problem to be solved in the present invention is to provide a kind of lithium battery cascade utilization screening method, to solve prior art
Existing conventional contact measurement lithium battery capacity and internal resistance parameter can cause secondary loss and screening efficiency low to lithium battery
The problem of.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of lithium battery cascade utilization screening method, including:
Obtain training sample data and training sample label;Wherein, the training sample data are the sample for training
The CT images of lithium battery, the training sample label includes:The corresponding label information of CT images, the label information is used to identify
Whether the corresponding sample lithium battery for being used to train being capable of echelon utilization;
It is trained, is instructed according to the disaggregated model that the training sample data and training sample label of acquisition treat training
Disaggregated model after white silk;
The CT images of lithium battery to be measured are obtained, the CT images of the lithium battery to be measured of acquisition are input to the classification mould after training
In type, whether the disaggregated model after training exports the lithium battery to be measured being capable of echelon utilization.
Further, the acquisition training sample label includes:
Measure capacity and the internal resistance of sample lithium battery for training;
Capacity is ranged less than the first predetermined threshold value and internal resistance higher than the sample lithium battery of the second predetermined threshold value can not ladder
The secondary lithium battery utilized, other sample lithium batteries are ranged can the lithium battery that utilizes of echelon.
Further, the disaggregated model for treating training in the training sample data and training sample label according to acquisition is carried out
Before training, the disaggregated model after being trained, methods described also includes:
Training sample data to acquisition carry out image procossing, and the contrast value for calculating training sample data is used as feature
Value;
The characteristic value of obtained training sample data will be calculated as the input of disaggregated model to be determined.
Further, the contrast value includes:Per pixel contrast, weber contrast, root mean square contrast, Michael
Inferior contrast.
Further, the training sample data and training sample label according to acquisition are treated the disaggregated model of training and entered
Row training, the disaggregated model after being trained includes:
The characteristic value of obtained training sample data will be calculated as the input of disaggregated model to be determined, the training obtained
Sample label as disaggregated model to be trained output;
The disaggregated model for treating training using the neural metwork training technology in supervised learning is trained, and is built based on branch
Hold the disaggregated model of vector machine algorithm.
Further, the disaggregated model after the training is categorised decision plane;
The CT images of the lithium battery to be measured by acquisition are input in the disaggregated model after training, the classification after training
The classification that model exports the lithium battery to be measured includes:
The CT images of the lithium battery to be measured of acquisition are input in the disaggregated model after training, according to the lithium to be measured of input electricity
The characteristic value in pond is located at the which side of categorised decision plane, determines whether lithium battery to be measured being capable of echelon utilization.
Further, the disaggregated model for treating training in the training sample data and training sample label according to acquisition is carried out
After training, the disaggregated model after being trained, methods described also includes:
Obtain test sample data and test sample label;Wherein, the test sample data are the sample for test
The CT images of lithium battery, the test sample label includes:The corresponding label information of CT images, the label information is used to identify
Whether the corresponding sample lithium battery for being used to test being capable of echelon utilization;
The contrast value for calculating test sample data is used as characteristic value;
The characteristic value of obtained test sample data will be calculated as the input of the disaggregated model after training;
The prediction label of the disaggregated model output after training is carried out into matching with the corresponding test sample label obtained to test
Card.
Further, it is trained in the disaggregated model that training is treated with the neural metwork training technology in machine learning,
Build after the disaggregated model based on algorithm of support vector machine, methods described also includes:
Using genetic algorithm, the parameter in algorithm of support vector machine is optimized, wherein, the parameter of optimization includes:Core
Function parameter and error penalty coefficient.
The above-mentioned technical proposal of the present invention has the beneficial effect that:
In such scheme, training sample data and training sample label are obtained;Wherein, the training sample data be for
The CT images of the sample lithium battery of training, the training sample label includes:The corresponding label information of CT images, the label letter
Cease for identifying the corresponding sample lithium battery for being used to train whether being capable of echelon utilization;According to the training sample data of acquisition and
The disaggregated model that training sample label treats training is trained, the disaggregated model after being trained;Obtain lithium battery to be measured
CT images, the CT images of the lithium battery to be measured of acquisition are input in the disaggregated model after training, the disaggregated model after training
Whether export the lithium battery to be measured being capable of echelon utilization.So, when it is determined that whether lithium battery can be utilized by echelon, only need
The CT images of corresponding lithium battery are obtained, it is not necessary to obtain lithium battery parameter, so as to overcome conventional contact measurement lithium electricity
The deficiency of pond electrical parameter, belongs to lossless contactless screening technique, and this screening technique to lithium battery without carrying out discharge and recharge
Process, shortens screening time, without the internal resistance for detecting lithium battery one by one, reduces cost of labor, so as to improving sieve
Select efficiency.
Brief description of the drawings
Fig. 1 is the schematic flow sheet of lithium battery cascade utilization screening method provided in an embodiment of the present invention;
Fig. 2 is the detailed process schematic diagram of lithium battery cascade utilization screening method provided in an embodiment of the present invention;
Fig. 3 is the schematic flow sheet of machine learning screening technique provided in an embodiment of the present invention;
Embodiment
To make the technical problem to be solved in the present invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing and tool
Body embodiment is described in detail.
The present invention measures lithium battery capacity and internal resistance parameter for existing conventional contact to cause secondary to lithium battery
There is provided a kind of lithium battery cascade utilization screening method for the problem of loss and screening efficiency are low.
As shown in figure 1, lithium battery cascade utilization screening method provided in an embodiment of the present invention, including:
S101, obtains training sample data and training sample label;Wherein, the training sample data are for training
Computed tomography (Computed Tomography, CT) image of sample lithium battery, the training sample label includes:
Whether the corresponding label information of CT images, the label information is used to identify the corresponding sample lithium battery for being used to train being capable of ladder
It is secondary to utilize;
S102, is trained according to the disaggregated model that the training sample data and training sample label of acquisition treat training,
Disaggregated model after being trained;
S103, obtains the CT images of lithium battery to be measured, and the CT images of the lithium battery to be measured of acquisition are input to after training
In disaggregated model, whether the disaggregated model after training exports the lithium battery to be measured being capable of echelon utilization.
Lithium battery cascade utilization screening method described in the embodiment of the present invention, obtains training sample data and training sample mark
Label;Wherein, the training sample data are the CT images of the sample lithium battery for training, and the training sample label includes:
Whether the corresponding label information of CT images, the label information is used to identify the corresponding sample lithium battery for being used to train being capable of ladder
It is secondary to utilize;It is trained, is instructed according to the disaggregated model that the training sample data and training sample label of acquisition treat training
Disaggregated model after white silk;The CT images of lithium battery to be measured are obtained, the CT images of the lithium battery to be measured of acquisition are input to after training
Disaggregated model in, disaggregated model after training export the lithium battery to be measured whether can echelon utilize.So, it is determined that
When whether lithium battery can be utilized by echelon, it is only necessary to obtain the CT images of corresponding lithium battery, it is not necessary to obtain lithium battery ginseng
Number, so as to overcome the deficiency that conventional contact measures lithium battery parameter, belongs to lossless contactless screening technique, and this
Screening technique shortens screening time without carrying out charge and discharge process to lithium battery, without the internal resistance for detecting lithium battery one by one,
Reduce cost of labor, so as to improving screening efficiency.
Lithium battery cascade utilization screening method described in the present embodiment as it is a kind of rapidly and efficiently, it is lossless contactless new
Type triage techniques, with potential commercial advantages, is adapted to the specialized taking back and process of enterprise.
In the present embodiment, with reference to lithium battery failure cause analysis, can the lithium battery that utilizes of echelon generally than it is discarded (can not ladder
It is secondary to utilize) lithium battery there is apparent internal structure, this clearly structural information can be reflected in the CT figures of lithium battery
As in.
In the present embodiment, in order to obtain disaggregated model, need to obtain training sample data and training sample label first, its
In, the training sample data are the CT images of the sample lithium battery for training, and the training sample label includes:CT images
Whether corresponding label information, the label information is used to identify the corresponding sample lithium battery for being used to train being capable of echelon profit
With for identifying sample lithium battery whether being capable of the label information that utilizes of echelon, the sample lithium battery that can be obtained according to measurement
Capacity and internal resistance determine.
In the present embodiment, the sample lithium battery for training includes but is not limited to:Waste lithium cell, for example or newly
Lithium battery;In the present embodiment, the CT images to acquisition for the sample lithium battery of training and acquisition by taking waste lithium cell as an example
Specific steps for the label information of the sample lithium battery of training are illustrated:
Certain amount waste lithium cell is collected as the sample lithium battery for training, these lithium batteries are swept using optics
Retouch and obtain CT images, obtained CT images are as training sample data;Measure the electricity such as internal resistance, the capacity of these lithium batteries simultaneously
Parameter, evaluate each lithium battery whether can echelon by the use of being used as training sample label during machine learning.
Machine learning, which is divided into supervised learning and the major class of unsupervised learning two, the present embodiment, uses supervised learning, prison
Educational inspector, which practises, needs training sample data and training sample label, and the training sample data used are exactly to scan waste lithium cell acquisition
CT images;Training sample label refer to need to know in advance these be used for train sample lithium battery whether can echelon it is sharp
With, these whether can the information that utilizes of echelon be referred to as label, in specific implementation process, can represent lithium battery not with -1
Energy echelon is utilized, and 1 represents that lithium battery can echelon utilization.This is also to be called supervised learning why, and the meaning is exactly to have known in advance
Lithium battery whether can echelon utilize, regard training sample data as the input of disaggregated model to be trained, training sample label
As the output of disaggregated model to be trained, disaggregated model to be trained is trained with training sample data, then again with training
Disaggregated model, to do not know whether can echelon using waste lithium cell classify.
It is further, described to obtain training sample in the embodiment of foregoing lithium battery cascade utilization screening method
This label includes:
Measure capacity and the internal resistance of sample lithium battery for training;
Capacity is ranged less than the first predetermined threshold value and internal resistance higher than the sample lithium battery of the second predetermined threshold value can not ladder
The secondary lithium battery utilized, other sample lithium batteries are ranged can the lithium battery that utilizes of echelon.
In the present embodiment, traditional lithium battery classification is to measure internal resistance and capacity, internal resistance and appearance one by one to lithium battery
The measurement of amount is required for measuring apparatus and lithium battery directly to contact, and the lithium battery cascade utilization screening method described in the present embodiment is only
A number of training sample data (capacity and internal resistance information that are not required to measure lithium battery) and training sample label is needed (to need to survey
Measure the capacity and internal resistance information of lithium battery) set up disaggregated model.
After disaggregated model is established, it is not necessary to which the capacity and internal resistance information for measuring lithium battery just can be by waste lithium cells point
Class, so as to overcome the deficiency that conventional contact measures lithium battery parameter, the lithium battery echelon described in the present embodiment utilizes sieve
Choosing method belongs to lossless contactless screening technique.
In the present embodiment, the capacity and internal resistance information of lithium battery are measured, is to obtain the sample lithium battery for training
Label information, i.e.,:The corresponding lithium battery of every width CT images whether can echelon utilize.
In the present embodiment, echelon utilization is unable to for example, capacity can be less than to 10% and ranged with internal resistance higher than 200m Ω
Lithium battery, other are ranged can the lithium battery that utilizes of echelon.
In the embodiment of foregoing lithium battery cascade utilization screening method, further, in the instruction according to acquisition
The disaggregated model that white silk sample data and training sample label treat training is trained, before the disaggregated model after being trained,
Methods described also includes:
Training sample data to acquisition carry out image procossing, and the contrast value for calculating training sample data is used as feature
Value;
The characteristic value of obtained training sample data will be calculated as the input of disaggregated model to be determined.
In the present embodiment, according to can echelon utilize and be unable to the lithium battery that echelon is utilized CT images significant difference,
The CT images of the sample lithium battery for being used to train of acquisition can be handled, the contrast value for calculating training sample data is made
It is characterized value;The characteristic value of obtained training sample data will be calculated as the input of disaggregated model to be determined.
In the embodiment of foregoing lithium battery cascade utilization screening method, further, the contrast value bag
Include:Per pixel contrast, weber contrast, root mean square contrast, Michelson contrast.
In the present embodiment, the contrast value can include:Per pixel contrast, weber contrast, root mean square contrast,
Michelson contrast, the characteristic vector that these contrast values can be recognized as intelligent classification.
It is further, described according to acquisition in the embodiment of foregoing lithium battery cascade utilization screening method
The disaggregated model that training sample data and training sample label treat training is trained, the disaggregated model bag after being trained
Include:
The characteristic value of obtained training sample data will be calculated as the input of disaggregated model to be determined, the training obtained
Sample label as disaggregated model to be trained output;
The disaggregated model for treating training using the neural metwork training technology in supervised learning is trained, and is built based on branch
Hold the disaggregated model of vector machine algorithm.
In the present embodiment, supervised learning be exactly with it is known whether can the training sample data that utilize of echelon remove training nerve
Network obtains disaggregated model with this, specifically:According to algorithm of support vector machine, the feature of obtained training sample data will be calculated
Value is used as the output of disaggregated model to be trained, profit as the training sample label of input, the acquisition of disaggregated model to be determined
The disaggregated model for treating training with the neural metwork training technology in supervised learning is trained, and training terminates resulting classification
Model is the disaggregated model of requirement.
In the present embodiment, using the method for supervised learning, the learning ability of computer disposal mass data and quick is utilized
Accurate computing capability, reaches and fast and accurately screens purpose to waste lithium cell.
In the embodiment of foregoing lithium battery cascade utilization screening method, further, point after the training
Class model is categorised decision plane;
The CT images of the lithium battery to be measured by acquisition are input in the disaggregated model after training, the classification after training
The classification that model exports the lithium battery to be measured includes:
The CT images of the lithium battery to be measured of acquisition are input in the disaggregated model after training, according to the lithium to be measured of input electricity
The characteristic value in pond is located at the which side of categorised decision plane, determines whether lithium battery to be measured being capable of echelon utilization.
In the present embodiment, the disaggregated model obtained after the training is substantially a categorised decision plane, will can be treated
Survey lithium battery according to it is no can echelon using separating, specifically:The CT images of the lithium battery to be measured of acquisition are input to instruction
In disaggregated model after white silk, the which side of categorised decision plane is located at according to the characteristic value of the lithium battery to be measured of input, it is determined that treating
Whether survey lithium battery being capable of echelon utilization.
In the embodiment of foregoing lithium battery cascade utilization screening method, further, in the instruction according to acquisition
The disaggregated model that white silk sample data and training sample label treat training is trained, after the disaggregated model after being trained,
Methods described also includes:
Obtain test sample data and test sample label;Wherein, the test sample data are the sample for test
The CT images of lithium battery, the test sample label includes:The corresponding label information of CT images, the label information is used to identify
Whether the corresponding sample lithium battery for being used to test being capable of echelon utilization;
The contrast value for calculating test sample data is used as characteristic value;
The characteristic value of obtained test sample data will be calculated as the input of the disaggregated model after training;
The prediction label of the disaggregated model output after training is carried out into matching with the corresponding test sample label obtained to test
Card.
In the present embodiment, the sample lithium battery for test includes but is not limited to:Waste lithium cell, for example or newly
Lithium battery.
In the embodiment of foregoing lithium battery cascade utilization screening method, further, with machine learning
Neural metwork training technology treat the disaggregated model of training and be trained, build the disaggregated model based on algorithm of support vector machine
Afterwards, methods described also includes:
Using genetic algorithm, the parameter in algorithm of support vector machine is optimized, wherein, the parameter of optimization includes:Core
Function parameter and error penalty coefficient.
In the present embodiment, the parameter of obtained disaggregated model can also be optimized, specifically:Using genetic algorithm,
Kernel functional parameter in algorithm of support vector machine and error penalty coefficient are optimized, the accuracy rate of sifting sort is improved, makes
The discrimination for obtaining disaggregated model can be close to 90%.
Lithium battery cascade utilization screening method described in embodiment for a better understanding of the present invention, to the embodiment of the present invention
Described lithium battery cascade utilization screening method is described in detail, as shown in Figures 2 and 3, the lithium described in the embodiment of the present invention
Battery cascade utilization screening method specifically may comprise steps of:
Step 1, acquisition sample data (training sample data and test sample data) and sample label (training sample label
With test sample label):
1. collecting 18650 model waste lithium cells of 200 different manufacturers as sample lithium battery, optical scanning device is used
The CT images of each lithium battery are obtained, the CT images of 200 lithium batteries are obtained as sample data.
2. measure the capacity of this 200 lithium batteries, internal resistance parameter one by one, by capacity less than 10% and internal resistance is higher than 200m Ω
Lithium battery be recorded as being unable to the lithium battery that echelon is utilized, represented with -1, other be recorded as can the lithium battery that utilizes of echelon, with 1
Represent, obtain sample label.
Step 2, obtained CT images are handled, extract characteristic value information:
1.CT image preprocessings:
Because machine can only be operated to data, therefore, it is that can not know directly as machine to scan obtained CT images
Other content, can only choose some characteristic informations of CT images, be referred to as characteristic value.Because the calculating of characteristic value is needed to image
Picture element matrix is calculated, and the CT images obtained may have different degrees of white background, and calculating can be caused to do greatly very much
Disturb.Therefore the white background of sheet is removed, that is to say, that by CT images firstly the need of interference is removed in use
White background cut, in order to calculate the progress of characteristic value.
The program circuit for cutting white background is as follows:
CT images are deposited according to matrix form, the pixel of the point in each data record image in matrix, can
It is to read image with the gray scale value matrix that CT images are obtained by matlab softwares, the side of CT images is found in matlab softwares
Pixel dot position information, determines clipping boundary at boundary, record delimitation, is completed using function is cut.
2. calculate characteristic value:
Can the lithium battery that is generally utilized than being unable to echelon of the lithium battery that utilizes of echelon there is apparent internal structure, it is and right
It is the major parameter for weighing gray level image definition than degree.At present there are distinct methods in the definition for gray level image contrast,
Contrast under being defined from a variety of distinct methods is as characteristic value, as shown in figure 3, the contrast used in the present embodiment
Including:Per pixel contrast, weber contrast, root mean square contrast, Michelson contrast.
Then, the contrast under being defined to distinct methods is illustrated:
(1) per pixel contrast Cpp
The accumulative strength difference between current pixel and an adjacent pixel ring is defined as per pixel contrast:
Wherein, i=1:M, j=1:n;M, n are CT gradation of image image array sizes;I (i, j) is the i-th row jth row pixel
Gray value.
OrBorder condition under, skip I (x, y) calculating.
The average C of whole imagePPIt is
(2) weber contrast Cw
Weber contrast is defined as:
Wherein, IbIt is the intensity level of background, background is white in the present embodiment, therefore IbEqual to 255.Due to IbIt is always big
In or equal to I (i, j), therefore | I (i, j)-Ib| can be by Ib- I (i, j) is substituted.
The weber contrast of entire image is:
(3) Michelson contrast Cm
Michelson contrast is defined as:
Wherein, Imax(i, j)=arg maxx∈[i-1,i+1],y∈[j-1,j+1]I(x,y)
Imin(x, y)=arg minx∈[i-1,i+1],y∈[j-1,j+1]I(x,y)
The average Michelson contrast of whole image is:
(4) root mean square contrast is defined as:
Wherein,
The program circuit for calculating characteristic value is as follows:
CT gray level image matrixes are read, contrast calculation procedure are write according to different contrast definition, to sample lithium battery
Every piece image seek out 4 characteristic values according to the definition of 4 contrasts above, be stored in Microsoft Excel;And with 1, -1 table
Show sample label, wherein, 1 represent can the lithium battery that utilizes of echelon, -1 represents to be unable to the lithium battery that echelon is utilized, by sample data
Sample label information be stored in Microsoft Excel;So, a lithium battery just obtains one group of corresponding record:4 contrasts
Parameter and label information.
Step 3, using matlab algorithm of support vector machine program write according to SVMs theory, with wherein 150
As the training sample data and training sample label of train classification models, this 150 institutes are read from Microsoft Excel data right
The contrast level parameter and label information answered, are trained to disaggregated model, the disaggregated model after being trained.
Step 4, with the remaining 50 test sample data and test sample label as testing classification model, test instruction
The disaggregated model obtained after white silk, the prediction label of this 50 lithium batteries is exported by the disaggregated model obtained after training;Due to thing
First know the true tag of this 50 lithium batteries (i.e.:Test sample label), prediction label and true tag are matched, root
Classification and Identification rate is obtained according to matching degree;If, the now classification knowledge of the label of correctly predicted 40 width image, then this disaggregated model
Rate is not 80%.
Step 5, the parameter in algorithm of support vector machine is optimized, improve Classification and Identification effect.
Algorithm of support vector machine classifying quality depends primarily on kernel functional parameter gama i.e. γ and error penalty coefficient cost
That is C, disaggregated model effect can be greatly improved by finding optimized parameter.Using genetic algorithm to the gama in SVMs and
Cost parameters are optimized.Program circuit is as follows:
1) parameter of initialization SVMs (Support Vector Machine, SVM), influence SVMs point
The parameter of class effect mainly has two, i.e. kernel functional parameter gamma abbreviations γ, and error penalty coefficient Cost, abbreviation C, really
Determine the initial population of genetic algorithm, Population can choose 100, to SVMs need optimize two parameters (γ and
C) carry out binary coding and obtain 100 initial colonies, wherein, the length of binary coding sequence is according to the SVM to be searched for
The scope of parameter is determined with precision;
2) parameter of genetic algorithm, first initial algebra, initial crossover probability, initial mutation probability, maximum genetic algebra are set
Deng;
3) fitness that SVM is trained and calculates individual is sent into after initial population is decoded, fitness function uses SVM
The accuracy of classification;
4) optimum maintaining strategy is applied, first fitness (accuracy) highest individual is preserved before genetic manipulation is carried out
Get off, to prevent outstanding gene from, because genetic operator is operated and loses, recording worst individual sequence number Index;
5) genetic manipulation is carried out to above-mentioned initial population, specifically:Selection opertor uses roulette wheel selection method, crossover operator
Intersected using single-point, new colony is produced by genetic operator operation, and use fitness (accuracy) highest individual 4) preserved
Serial number Index new individual is replaced, and finally produces new colony, is then back to and 3) is trained;
6) check whether and meet algorithm end condition:Due to classification accuracy rate (fitness) knot inherently to be searched for
Really, it is difficult to as end condition, but when constant generations optimum individual fitness close to it is equal when then think that population can not be again
Evolve, algorithm is terminated;Or using the maximum genetic algebra of setting as algorithm end condition, when meeting in above-mentioned two condition
Any one when, then automatic termination algorithm.The later disaggregated model discrimination of optimization can be close to 90%.
It should be noted that herein, such as first and second or the like relational terms are used merely to a reality
Body or operation make a distinction with another entity or operation, and not necessarily require or imply these entities or deposited between operating
In any this actual relation or order.
Described above is the preferred embodiment of the present invention, it is noted that for those skilled in the art
For, on the premise of principle of the present invention is not departed from, some improvements and modifications can also be made, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (8)
1. a kind of lithium battery cascade utilization screening method, it is characterised in that including:
Obtain training sample data and training sample label;Wherein, the training sample data are the sample lithium electricity for training
The CT images in pond, the training sample label includes:The corresponding label information of CT images, the label information is used to identify accordingly
Be used for train sample lithium battery whether can echelon utilize;
It is trained, is obtained after training according to the disaggregated model that the training sample data and training sample label of acquisition treat training
Disaggregated model;
The CT images of lithium battery to be measured are obtained, the CT images of the lithium battery to be measured of acquisition are input to the disaggregated model after training
In, whether the disaggregated model after training exports the lithium battery to be measured being capable of echelon utilization.
2. lithium battery cascade utilization screening method according to claim 1, it is characterised in that the acquisition training sample mark
Label include:
Measure capacity and the internal resistance of sample lithium battery for training;
Capacity is ranged less than the first predetermined threshold value and internal resistance higher than the sample lithium battery of the second predetermined threshold value and is unable to echelon profit
Lithium battery, other sample lithium batteries are ranged can the lithium battery that utilizes of echelon.
3. lithium battery cascade utilization screening method according to claim 1, it is characterised in that in the training sample according to acquisition
The disaggregated model that notebook data and training sample label treat training is trained, described before the disaggregated model after being trained
Method also includes:
Training sample data to acquisition carry out image procossing, and the contrast value for calculating training sample data is used as characteristic value;
The characteristic value of obtained training sample data will be calculated as the input of disaggregated model to be determined.
4. lithium battery cascade utilization screening method according to claim 3, it is characterised in that the contrast value includes:
Per pixel contrast, weber contrast, root mean square contrast, Michelson contrast.
5. lithium battery cascade utilization screening method according to claim 3, it is characterised in that the training according to acquisition
The disaggregated model that sample data and training sample label treat training is trained, and the disaggregated model after being trained includes:
The characteristic value of obtained training sample data will be calculated as the input of disaggregated model to be determined, the training sample obtained
Label as disaggregated model to be trained output;
The disaggregated model for treating training using the neural metwork training technology in supervised learning is trained, build based on support to
The disaggregated model of amount machine algorithm.
6. lithium battery cascade utilization screening method according to claim 1, it is characterised in that the classification mould after the training
Type is categorised decision plane;
The CT images of the lithium battery to be measured by acquisition are input in the disaggregated model after training, the disaggregated model after training
Exporting the classification of the lithium battery to be measured includes:
The CT images of the lithium battery to be measured of acquisition are input in the disaggregated model after training, according to the lithium battery to be measured of input
Characteristic value is located at the which side of categorised decision plane, determines whether lithium battery to be measured being capable of echelon utilization.
7. lithium battery cascade utilization screening method according to claim 1, it is characterised in that in the training sample according to acquisition
The disaggregated model that notebook data and training sample label treat training is trained, described after the disaggregated model after being trained
Method also includes:
Obtain test sample data and test sample label;Wherein, the test sample data are the sample lithium electricity for test
The CT images in pond, the test sample label includes:The corresponding label information of CT images, the label information is used to identify accordingly
Be used for test sample lithium battery whether can echelon utilize;
The contrast value for calculating test sample data is used as characteristic value;
The characteristic value of obtained test sample data will be calculated as the input of the disaggregated model after training;
The prediction label of disaggregated model output after training is subjected to matching checking with the corresponding test sample label obtained.
8. lithium battery cascade utilization screening method according to claim 5, it is characterised in that with the god in machine learning
The disaggregated model for treating training through network training technology is trained, build the disaggregated model based on algorithm of support vector machine it
Afterwards, methods described also includes:
Using genetic algorithm, the parameter in algorithm of support vector machine is optimized, wherein, the parameter of optimization includes:Kernel function
Parameter and error penalty coefficient.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710565905.7A CN107330474A (en) | 2017-07-12 | 2017-07-12 | A kind of lithium battery cascade utilization screening method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710565905.7A CN107330474A (en) | 2017-07-12 | 2017-07-12 | A kind of lithium battery cascade utilization screening method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107330474A true CN107330474A (en) | 2017-11-07 |
Family
ID=60197594
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710565905.7A Pending CN107330474A (en) | 2017-07-12 | 2017-07-12 | A kind of lithium battery cascade utilization screening method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107330474A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647588A (en) * | 2018-04-24 | 2018-10-12 | 广州绿怡信息科技有限公司 | Goods categories recognition methods, device, computer equipment and storage medium |
CN109047040A (en) * | 2018-06-12 | 2018-12-21 | 东莞市德尔能新能源股份有限公司 | A kind of lithium battery screening system based on big data |
CN109604186A (en) * | 2018-12-14 | 2019-04-12 | 北京匠芯电池科技有限公司 | Power battery performance flexibility assesses method for separating |
CN110135490A (en) * | 2019-05-10 | 2019-08-16 | 上海理工大学 | A kind of retired lithium battery classification method based on sample label |
CN110336051A (en) * | 2019-06-05 | 2019-10-15 | 武汉理工大学 | The stage division that waste and old solid oxide fuel cell echelon utilizes |
CN112966835A (en) * | 2021-02-03 | 2021-06-15 | 上海电气集团股份有限公司 | Waste lithium battery recovery management system and method |
CN113466706A (en) * | 2021-07-26 | 2021-10-01 | 上海伟翔众翼新能源科技有限公司 | Lithium battery echelon utilization residual life prediction method based on convolutional neural network |
CN113996564A (en) * | 2021-12-02 | 2022-02-01 | 格林美股份有限公司 | Lithium battery echelon utilization and sorting method and device based on characteristic numerical analysis |
CN114047444A (en) * | 2021-11-09 | 2022-02-15 | 中国南方电网有限责任公司超高压输电公司广州局 | Storage battery health condition evaluation method and device |
CN114210591A (en) * | 2021-12-02 | 2022-03-22 | 格林美股份有限公司 | Lithium battery echelon utilization and sorting method and device based on IC curve |
CN114638834A (en) * | 2022-05-23 | 2022-06-17 | 深圳鑫茂新能源技术有限公司 | Waste lithium battery classification processing method based on image processing |
CN114669508A (en) * | 2022-03-01 | 2022-06-28 | 常州大学 | Screening method for graded utilization monomers of retired batteries |
CN115921356A (en) * | 2023-01-10 | 2023-04-07 | 北京凌禾科技有限公司 | Treatment method and treatment system for waste lithium batteries |
CN116401604A (en) * | 2019-05-13 | 2023-07-07 | 北京绪水互联科技有限公司 | Method for classifying and detecting cold head state and predicting service life |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103337671A (en) * | 2013-06-27 | 2013-10-02 | 国家电网公司 | Cascade utilization screening method of waste power batteries |
CN103901350A (en) * | 2014-03-13 | 2014-07-02 | 奇瑞汽车股份有限公司 | Worn-out power battery secondary use screening method |
CN104751447A (en) * | 2015-01-10 | 2015-07-01 | 哈尔滨工业大学(威海) | Lithium battery unit defect detection method |
CN105427335A (en) * | 2015-12-31 | 2016-03-23 | 先进储能材料国家工程研究中心有限责任公司 | Method for detecting and locating skip plating defects of continuous strip-shaped porous metal material |
-
2017
- 2017-07-12 CN CN201710565905.7A patent/CN107330474A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103337671A (en) * | 2013-06-27 | 2013-10-02 | 国家电网公司 | Cascade utilization screening method of waste power batteries |
CN103901350A (en) * | 2014-03-13 | 2014-07-02 | 奇瑞汽车股份有限公司 | Worn-out power battery secondary use screening method |
CN104751447A (en) * | 2015-01-10 | 2015-07-01 | 哈尔滨工业大学(威海) | Lithium battery unit defect detection method |
CN105427335A (en) * | 2015-12-31 | 2016-03-23 | 先进储能材料国家工程研究中心有限责任公司 | Method for detecting and locating skip plating defects of continuous strip-shaped porous metal material |
Non-Patent Citations (4)
Title |
---|
刘东平,单甘霖,张岐龙,段修生: "基于改进遗传算法的支持向量机参数优化", 《微计算机应用》 * |
周传兴等: "基于视觉感知的网络视频质量评价方法研究", 《微型机与应用》 * |
张宏军: "结合PCA和SVM的太阳能电池缺陷识别", 《电视技术》 * |
曹厚德: "图像质量评价与控制", 《现代医学影响技术学》 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647588A (en) * | 2018-04-24 | 2018-10-12 | 广州绿怡信息科技有限公司 | Goods categories recognition methods, device, computer equipment and storage medium |
CN109047040A (en) * | 2018-06-12 | 2018-12-21 | 东莞市德尔能新能源股份有限公司 | A kind of lithium battery screening system based on big data |
CN109604186B (en) * | 2018-12-14 | 2021-12-07 | 蓝谷智慧(北京)能源科技有限公司 | Flexible evaluation and sorting method for performance of power battery |
CN109604186A (en) * | 2018-12-14 | 2019-04-12 | 北京匠芯电池科技有限公司 | Power battery performance flexibility assesses method for separating |
CN110135490A (en) * | 2019-05-10 | 2019-08-16 | 上海理工大学 | A kind of retired lithium battery classification method based on sample label |
CN110135490B (en) * | 2019-05-10 | 2023-08-01 | 上海理工大学 | Retired lithium battery classification method based on sample label |
CN116401604B (en) * | 2019-05-13 | 2024-05-28 | 北京绪水互联科技有限公司 | Method for classifying and detecting cold head state and predicting service life |
CN116401604A (en) * | 2019-05-13 | 2023-07-07 | 北京绪水互联科技有限公司 | Method for classifying and detecting cold head state and predicting service life |
CN110336051B (en) * | 2019-06-05 | 2020-12-29 | 武汉理工大学 | Grading method for gradient utilization of waste solid oxide fuel cells |
CN110336051A (en) * | 2019-06-05 | 2019-10-15 | 武汉理工大学 | The stage division that waste and old solid oxide fuel cell echelon utilizes |
CN112966835A (en) * | 2021-02-03 | 2021-06-15 | 上海电气集团股份有限公司 | Waste lithium battery recovery management system and method |
CN112966835B (en) * | 2021-02-03 | 2023-12-29 | 上海电气集团股份有限公司 | Waste lithium battery recycling management system and method |
CN113466706A (en) * | 2021-07-26 | 2021-10-01 | 上海伟翔众翼新能源科技有限公司 | Lithium battery echelon utilization residual life prediction method based on convolutional neural network |
CN113466706B (en) * | 2021-07-26 | 2022-07-29 | 上海伟翔众翼新能源科技有限公司 | Lithium battery echelon utilization residual life prediction method based on convolutional neural network |
CN114047444A (en) * | 2021-11-09 | 2022-02-15 | 中国南方电网有限责任公司超高压输电公司广州局 | Storage battery health condition evaluation method and device |
CN114047444B (en) * | 2021-11-09 | 2024-05-28 | 中国南方电网有限责任公司超高压输电公司广州局 | Method and device for evaluating health condition of storage battery |
CN114210591B (en) * | 2021-12-02 | 2023-12-22 | 格林美股份有限公司 | Lithium battery echelon utilization sorting method and device based on IC curve |
CN114210591A (en) * | 2021-12-02 | 2022-03-22 | 格林美股份有限公司 | Lithium battery echelon utilization and sorting method and device based on IC curve |
CN113996564A (en) * | 2021-12-02 | 2022-02-01 | 格林美股份有限公司 | Lithium battery echelon utilization and sorting method and device based on characteristic numerical analysis |
CN114669508A (en) * | 2022-03-01 | 2022-06-28 | 常州大学 | Screening method for graded utilization monomers of retired batteries |
CN114638834A (en) * | 2022-05-23 | 2022-06-17 | 深圳鑫茂新能源技术有限公司 | Waste lithium battery classification processing method based on image processing |
CN115921356A (en) * | 2023-01-10 | 2023-04-07 | 北京凌禾科技有限公司 | Treatment method and treatment system for waste lithium batteries |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107330474A (en) | A kind of lithium battery cascade utilization screening method | |
CN110224192B (en) | Method for predicting service life of power battery by gradient utilization | |
CN113158947B (en) | Power battery health scoring method, system and storage medium | |
Zhou et al. | An efficient screening method for retired lithium-ion batteries based on support vector machine | |
CN109604186B (en) | Flexible evaluation and sorting method for performance of power battery | |
CN110068774A (en) | Estimation method, device and the storage medium of lithium battery health status | |
CN110263934B (en) | Artificial intelligence data labeling method and device | |
CN108334900B (en) | Generation method and system of classification model of power battery, and classification method and system | |
CN106651574A (en) | Personal credit assessment method and apparatus | |
CN112379269A (en) | Battery abnormity detection model training and detection method and device thereof | |
CN112287980B (en) | Power battery screening method based on typical feature vector | |
CN114707571B (en) | Credit data anomaly detection method based on enhanced isolation forest | |
CN114280479A (en) | Electrochemical impedance spectrum-based rapid sorting method for retired batteries | |
CN107704883A (en) | A kind of sorting technique and system of the grade of magnesite ore | |
CN111639882A (en) | Deep learning-based power utilization risk judgment method | |
CN112686380A (en) | Neural network-based echelon power cell consistency evaluation method and system | |
CN115327417A (en) | Early warning method and system for abnormity of power battery monomer and electronic equipment | |
CN116150572A (en) | Automobile battery monomer consistency defect evaluation method based on cluster analysis | |
US20150242676A1 (en) | Method for the Supervised Classification of Cells Included in Microscopy Images | |
CN114460481A (en) | Energy storage battery thermal runaway early warning method based on Bi-LSTM and attention mechanism | |
CN112183459B (en) | Remote sensing water quality image classification method based on evolution multi-objective optimization | |
CN112906672A (en) | Steel rail defect identification method and system | |
CN117251814A (en) | Method for analyzing electric quantity loss abnormality of highway charging pile | |
CN117283372A (en) | Cutter wear monitoring method based on twin long-time memory neural network | |
CN112287979A (en) | Mutual information-based energy storage battery state judgment method |
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 |
Application publication date: 20171107 |
|
RJ01 | Rejection of invention patent application after publication |