CN107330474A - A kind of lithium battery cascade utilization screening method - Google Patents

Abstract

The invention provides a screening method for cascaded utilization of lithium batteries, which belongs to a non-destructive and non-contact screening method and can improve screening efficiency. The method includes: acquiring training sample data and training sample labels; wherein, the training sample data is a CT image of a sample lithium battery used for training, and the training sample label includes: label information corresponding to the CT image, the label The information is used to identify whether the corresponding sample lithium battery used for training can be used in stages; train the classification model to be trained according to the acquired training sample data and training sample labels, and obtain the trained classification model; obtain the CT of the lithium battery to be tested image, the acquired CT image of the lithium battery to be tested is input into the trained classification model, and the trained classification model outputs whether the lithium battery to be tested can be used in stages. The invention relates to the technical field of lithium battery recycling.

Description

A screening method for cascaded utilization of lithium batteries

technical field

The invention relates to the technical field of lithium battery recycling, in particular to a screening method for cascade utilization of lithium batteries.

Background technique

In recent years, the development speed of my country's electric vehicle industry will usher in the peak of power lithium battery retirement in the next two to three years, and the total amount of these lithium batteries will increase at an accelerated rate every year; by 2020, the stock of electric vehicle market will exceed 500 It is estimated that a car is equipped with an average of 20kWh of lithium batteries, and about 100 million kWh (1000GWh) of lithium-ion lithium batteries have entered the automotive market. As we all know, the chemical substances and heavy metal elements in lithium-ion lithium batteries will cause pollution and harm to the environment. How to maximize the recovery and cascade utilization of the remaining value of these lithium batteries is an urgent and very important scientific and technological issue.

For decommissioned power lithium batteries, there are two feasible treatment methods. One is to scrap and dismantle them directly as industrial waste products, extract the raw materials in them, and realize the recycling of raw materials. Some domestic enterprises are already doing this. Commercial operation; Another way is to consider the decommissioned power lithium battery. Although it has not met the conditions of use of the car, it still has a certain amount of surplus energy, and its life has not completely expired. It can be used in other fields as a carrier of electric energy. use, so as to give full play to its residual value. Obviously, the latter step-by-step utilization can give full play to the maximum value of products and maximize the benefits of circular economy, which is a greener and more environmentally friendly way.

Chinese invention patent CN201310261893.0 proposes a screening method for cascaded utilization of waste power lithium batteries: (1) Charge the waste power lithium battery pack so that the state of charge SOC is 15% to 80%; then disassemble the lithium battery pack , check the appearance of the power lithium battery pack and the single lithium battery, and record; (2) Detect and record the open circuit voltage and internal resistance of each single lithium battery, and the open circuit voltage and internal resistance of the standard single lithium battery Contrast; from the tested voltage and internal resistance according to the charging and discharging curve of the standard single lithium battery, evaluate the capacity of the single lithium battery of the waste power lithium battery pack; (3) connect the above-mentioned waste power lithium battery monomers in parallel until their open circuit voltage is basically the same , compared with the open circuit voltage of the single lithium battery before parallel connection, and record the voltage rise and fall; then put the waste single lithium battery and standard single lithium battery at a temperature of 30°C to 55°C for 3 to 7 days or at room temperature Put it aside for 10 to 30 days, check its open circuit voltage and internal resistance, and record it; conduct a discharge cycle test on a standard single lithium battery, and use the state of charge, capacity-voltage curve, and internal resistance of the standard single lithium battery as a reference, according to the waste power The open circuit voltage and internal resistance of lithium battery cells are used to evaluate the health status of waste power lithium battery cells; (4) According to the above records, compare the appearance, open circuit voltage, internal resistance, voltage drop and health status of waste power lithium battery cells , Classify the waste power lithium battery cells, and the lithium batteries of the same level are used in groups with the energy storage grid.

In the invention with the patent number CN 103901350 A, it involves a screening method for the secondary use of waste power lithium batteries: first, charge and discharge the lithium battery pack as a whole, and pass the data recorded by the BMS during the test. During the discharge process, select 2 -4 SOC points, the SOC value is between 20% and 90%, read the voltage of each series-connected monomer; identify the lithium battery that deviates from the voltage value of most lithium batteries by more than 5% as a problematic lithium battery, The remaining lithium batteries are initially identified as healthy lithium batteries; the internal resistance of each remaining lithium battery is measured, and the second screening is carried out through the internal resistance value, and the batteries that deviate from the normal internal resistance of most batteries by 20% are removed, and the screening is over. ; Among the remaining lithium batteries, 10 lithium batteries are randomly selected to be charged and discharged once, and the average capacity of the 10 lithium batteries is defaulted to the capacity value of all healthy lithium batteries.

The screening methods used in the prior art all more or less involve the measurement of the electrical parameters of the lithium battery, that is, the contact type, such as the detection of the open circuit voltage and internal resistance, which requires the process of charging and discharging the battery; on the one hand, the detection of these parameters The measurement may cause changes and damage to the internal structure of the lithium battery. On the other hand, the charging and discharging process of the lithium battery takes several hours to complete, resulting in a relatively high cost of time and labor, which leads to the recycling of commercial lithium batteries. The economic benefits associated with cascade utilization are limited.

Contents of the invention

The technical problem to be solved by the present invention is to provide a screening method for cascaded utilization of lithium batteries to solve the problem that the traditional contact measurement of lithium battery capacity and internal resistance parameters in the prior art will cause secondary loss of lithium batteries and low screening efficiency. question.

In order to solve the above technical problems, an embodiment of the present invention provides a lithium battery cascade utilization screening method, including:

Obtain training sample data and training sample labels; wherein, the training sample data is a CT image of a sample lithium battery used for training, and the training sample label includes: label information corresponding to the CT image, and the label information is used to identify the corresponding Whether the sample lithium battery used for training can be used in stages;

Training the classification model to be trained according to the acquired training sample data and the training sample label to obtain the trained classification model;

Obtain a CT image of the lithium battery to be tested, input the acquired CT image of the lithium battery to be tested into a trained classification model, and output whether the lithium battery to be tested can be used in stages by the trained classification model.

Further, said obtaining training sample labels includes:

Measure the capacity and internal resistance of the sample lithium battery used for training;

The sample lithium batteries whose capacity is lower than the first preset threshold and whose internal resistance is higher than the second preset threshold are classified as lithium batteries that cannot be used in steps, and other sample lithium batteries are classified as lithium batteries that can be used in steps.

Further, before training the classification model to be trained according to the acquired training sample data and training sample labels to obtain the trained classification model, the method also includes:

Perform image processing on the acquired training sample data, and calculate the contrast value of the training sample data as a feature value;

The calculated eigenvalues of the training sample data are used as the input of the classification model to be determined.

Further, the contrast value includes: per-pixel contrast, Weber contrast, root mean square contrast, and Michelson contrast.

Further, the classification model to be trained is trained according to the obtained training sample data and training sample labels, and the trained classification model includes:

Using the calculated eigenvalues of the training sample data as the input of the classification model to be determined, and the obtained training sample label as the output of the classification model to be trained;

The neural network training technology in supervised learning is used to train the classification model to be trained, and the classification model based on the support vector machine algorithm is constructed.

Further, the trained classification model is a classification decision plane;

The CT image of the lithium battery to be tested is input into the trained classification model, and the classification of the lithium battery to be tested is output by the trained classification model including:

Input the obtained CT image of the lithium battery to be tested into the trained classification model, and determine whether the lithium battery to be tested can be used step by step according to which side of the classification decision plane the input feature value of the lithium battery to be tested is located.

Further, after training the classification model to be trained according to the obtained training sample data and training sample labels, and obtaining the trained classification model, the method further includes:

Obtain test sample data and test sample labels; wherein, the test sample data is a CT image of a sample lithium battery used for testing, and the test sample label includes: label information corresponding to the CT image, and the label information is used to identify the corresponding Whether the sample lithium batteries used for testing can be used in stages;

Calculate the contrast value of the test sample data as the feature value;

The calculated eigenvalues of the test sample data are used as the input of the trained classification model;

The predicted label output by the trained classification model is matched with the obtained corresponding test sample label for verification.

Further, after using the neural network training technique in machine learning to train the classification model to be trained, after constructing the classification model based on the support vector machine algorithm, the method also includes:

The genetic algorithm is used to optimize the parameters in the support vector machine algorithm, wherein the optimized parameters include: kernel function parameters and error penalty coefficients.

The beneficial effects of above-mentioned technical scheme of the present invention are as follows:

In the above scheme, the training sample data and the training sample label are obtained; wherein, the training sample data is a CT image of a sample lithium battery used for training, and the training sample label includes: label information corresponding to the CT image, and the label information Used to identify whether the corresponding sample lithium battery used for training can be used in stages; train the classification model to be trained according to the acquired training sample data and training sample labels, and obtain the trained classification model; obtain the CT image of the lithium battery to be tested , inputting the obtained CT image of the lithium battery to be tested into the trained classification model, and outputting whether the lithium battery to be tested can be used step by step by the trained classification model. In this way, when determining whether a lithium battery can be used step by step, it is only necessary to obtain the CT image of the corresponding lithium battery, and it is not necessary to obtain the electrical parameters of the lithium battery, thus overcoming the shortcomings of the traditional contact method for measuring the electrical parameters of the lithium battery, and it is a non-destructive non-contact method. A screening method, and this screening method does not need to charge and discharge lithium batteries, shortens the screening time, and does not need to detect the internal resistance of lithium batteries one by one, reduces labor costs, thereby improving screening efficiency.

Description of drawings

Fig. 1 is a schematic flow chart of the step-by-step utilization screening method for lithium batteries provided by the embodiment of the present invention;

Fig. 2 is a detailed schematic flow diagram of the step-by-step utilization screening method for lithium batteries provided by the embodiment of the present invention;

FIG. 3 is a schematic flow chart of a machine learning screening method provided by an embodiment of the present invention;

detailed description

In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will describe in detail with reference to the drawings and specific embodiments.

Aiming at the problems that the existing traditional contact-type measurement of lithium battery capacity and internal resistance parameters will cause secondary loss and low screening efficiency of the lithium battery, the invention provides a lithium battery cascade utilization screening method.

As shown in Figure 1, the lithium battery cascade utilization screening method provided by the embodiment of the present invention includes:

S101, acquiring training sample data and training sample labels; wherein, the training sample data is a computed tomography (Computed Tomography, CT) image of a sample lithium battery used for training, and the training sample labels include: a label corresponding to the CT image Information, the label information is used to identify whether the corresponding sample lithium battery used for training can be used in stages;

S102. Train the classification model to be trained according to the acquired training sample data and the training sample label to obtain the trained classification model;

S103, acquiring a CT image of the lithium battery to be tested, inputting the acquired CT image of the lithium battery to be tested into a trained classification model, and outputting whether the lithium battery to be tested can be used step by step through the trained classification model.

The lithium battery step utilization screening method described in the embodiment of the present invention acquires training sample data and training sample labels; wherein, the training sample data is a CT image of a sample lithium battery used for training, and the training sample labels include: CT The label information corresponding to the image, the label information is used to identify whether the corresponding sample lithium battery used for training can be used in stages; according to the acquired training sample data and training sample labels, the classification model to be trained is trained to obtain the trained classification Model; obtain the CT image of the lithium battery to be tested, input the obtained CT image of the lithium battery to be tested into the trained classification model, and output whether the lithium battery to be tested can be used in stages by the trained classification model. In this way, when determining whether a lithium battery can be used step by step, it is only necessary to obtain the CT image of the corresponding lithium battery, and it is not necessary to obtain the electrical parameters of the lithium battery, thus overcoming the shortcomings of the traditional contact method for measuring the electrical parameters of the lithium battery, and it is a non-destructive non-contact method. A screening method, and this screening method does not need to charge and discharge lithium batteries, shortens the screening time, and does not need to detect the internal resistance of lithium batteries one by one, reduces labor costs, thereby improving screening efficiency.

The lithium battery cascade utilization screening method described in this example is a fast, efficient, non-destructive and non-contact new screening technology, which has potential commercial advantages and is suitable for professional recycling and treatment of enterprises.

In this embodiment, combined with the analysis of the failure causes of lithium batteries, lithium batteries that can be used in stages usually have a clearer internal structure than lithium batteries that are discarded (cannot be used in stages), and this clear structural information can be reflected in CT images of lithium batteries middle.

In this embodiment, in order to obtain the classification model, it is first necessary to obtain training sample data and training sample labels, wherein the training sample data is a CT image of a sample lithium battery used for training, and the training sample labels include: CT images corresponding to label information, the label information is used to identify whether the corresponding sample lithium battery used for training can be used in steps, and the label information used to identify whether the sample lithium battery can be used in steps can be obtained according to the measured capacity of the sample lithium battery and The internal resistance is determined.

In this embodiment, the sample lithium batteries used for training include but are not limited to: waste lithium batteries, for example, can also be new lithium batteries; The CT image and the specific steps to obtain the label information of the sample lithium battery used for training are described:

Collect a certain number of waste lithium batteries as sample lithium batteries for training, use optical scanning to obtain CT images of these lithium batteries, and obtain CT images as training sample data; at the same time measure the internal resistance, capacity and other electrical parameters of these lithium batteries, Evaluate whether each lithium battery can be used step by step as a training sample label for machine learning.

Machine learning is divided into two categories: supervised learning and unsupervised learning. In this embodiment, supervised learning is used. Supervised learning requires training sample data and training sample labels. The training sample data used is CT images obtained by scanning waste lithium batteries; The training sample label means that it is necessary to know in advance whether the lithium batteries used for training can be used in stages. The information on whether the lithium batteries can be used in stages is called a label. In the specific implementation process, -1 can be used to indicate that the lithium batteries cannot be used in stages. 1 indicates that the lithium battery can be used in stages. This is why it is called supervised learning, which means that it is known in advance whether lithium batteries can be used in stages, and the training sample data is used as the input of the classification model to be trained, and the training sample label is used as the output of the classification model to be trained. The trained classification model, and then use the trained classification model to classify the used lithium batteries that do not know whether they can be used in stages.

In the specific implementation of the aforementioned lithium battery cascade utilization screening method, further, said obtaining training sample labels includes:

Measure the capacity and internal resistance of the sample lithium battery used for training;

The sample lithium batteries whose capacity is lower than the first preset threshold and whose internal resistance is higher than the second preset threshold are classified as lithium batteries that cannot be used in steps, and other sample lithium batteries are classified as lithium batteries that can be used in steps.

In this embodiment, the traditional classification of lithium batteries is to measure the internal resistance and capacity of lithium batteries one by one. The measurement of internal resistance and capacity requires direct contact between the measuring equipment and lithium batteries. The step-by-step utilization screening method for lithium batteries described in this embodiment Only a certain amount of training sample data (no need to measure the capacity and internal resistance information of the lithium battery) and training sample labels (need to measure the capacity and internal resistance information of the lithium battery) are required to establish a classification model.

After the classification model is established, waste lithium batteries can be classified without measuring the capacity and internal resistance information of lithium batteries, thereby overcoming the shortcomings of traditional contact measurement of lithium battery electrical parameters. The step-by-step utilization screening of lithium batteries described in this embodiment The method belongs to the non-destructive non-contact screening method.

In this embodiment, the purpose of measuring the capacity and internal resistance information of the lithium battery is to obtain the label information of the sample lithium battery used for training, that is, whether the lithium battery corresponding to each CT image can be used in stages.

In this embodiment, for example, lithium batteries with capacities lower than 10% and internal resistances higher than 200 mΩ can be classified as lithium batteries that cannot be used in cascades, and others can be classified as lithium batteries that can be used in cascades.

In the specific implementation of the foregoing lithium battery cascade utilization screening method, further, before training the classification model to be trained according to the acquired training sample data and training sample labels to obtain the trained classification model, the method further includes:

Perform image processing on the acquired training sample data, and calculate the contrast value of the training sample data as a feature value;

The calculated eigenvalues of the training sample data are used as the input of the classification model to be determined.

In this embodiment, according to the obvious difference between the CT images of lithium batteries that can be used in steps and those that cannot be used in steps, the CT images of the sample lithium batteries acquired for training can be processed, and the contrast value of the training sample data can be calculated as the feature Value; use the calculated eigenvalues of the training sample data as the input of the classification model to be determined.

In the specific implementation of the foregoing lithium battery step utilization screening method, further, the contrast value includes: per-pixel contrast, Weber contrast, root mean square contrast, and Michelson contrast.

In this embodiment, the contrast value may include: per-pixel contrast, Weber contrast, root mean square contrast, and Michelson contrast, and these contrast values may be used as feature vectors for intelligent classification and identification.

In the specific implementation of the aforementioned lithium battery cascade utilization screening method, further, the classification model to be trained is trained according to the acquired training sample data and training sample labels, and the trained classification model includes:

Using the calculated eigenvalues of the training sample data as the input of the classification model to be determined, and the obtained training sample label as the output of the classification model to be trained;

The neural network training technology in supervised learning is used to train the classification model to be trained, and the classification model based on the support vector machine algorithm is constructed.

In this embodiment, supervised learning is to use the training sample data known whether it can be used step by step to train the neural network to obtain the classification model. Specifically: according to the support vector machine algorithm, the calculated eigenvalues of the training sample data are used as the The input of the determined classification model and the obtained training sample labels are used as the output of the classification model to be trained, and the classification model to be trained is trained using the neural network training technology in supervised learning. The classification model obtained after the training is the required classification Model.

In this embodiment, the method of supervised learning is adopted, and the learning ability and fast and accurate calculation ability of the computer to process massive data are used to achieve the purpose of fast and accurate screening of waste lithium batteries.

In the specific implementation of the aforementioned lithium battery cascade utilization screening method, further, the trained classification model is a classification decision plane;

The CT image of the lithium battery to be tested is input into the trained classification model, and the classification of the lithium battery to be tested is output by the trained classification model including:

Input the obtained CT image of the lithium battery to be tested into the trained classification model, and determine whether the lithium battery to be tested can be used step by step according to which side of the classification decision plane the input feature value of the lithium battery to be tested is located.

In this embodiment, the classification model obtained after the training is essentially a classification decision plane, which can separate the lithium battery to be tested according to whether it can be used in steps, specifically: input the acquired CT image of the lithium battery to be tested In the classification model after training, according to which side of the classification decision plane the input feature value of the lithium battery to be tested is located, it is determined whether the lithium battery to be tested can be used in stages.

In the specific implementation of the foregoing lithium battery cascade utilization screening method, further, after training the classification model to be trained according to the acquired training sample data and training sample labels, and obtaining the trained classification model, the method further includes:

Obtain test sample data and test sample labels; wherein, the test sample data is a CT image of a sample lithium battery used for testing, and the test sample label includes: label information corresponding to the CT image, and the label information is used to identify the corresponding Whether the sample lithium batteries used for testing can be used in stages;

Calculate the contrast value of the test sample data as the feature value;

The calculated eigenvalues of the test sample data are used as the input of the trained classification model;

The predicted label output by the trained classification model is matched with the obtained corresponding test sample label for verification.

In this embodiment, the sample lithium batteries used for testing include but are not limited to: waste lithium batteries, for example, new lithium batteries are also possible.

In the specific implementation of the foregoing lithium battery cascade utilization screening method, further, after using the neural network training technology in machine learning to train the classification model to be trained, and constructing the classification model based on the support vector machine algorithm, the method also include:

The genetic algorithm is used to optimize the parameters in the support vector machine algorithm, wherein the optimized parameters include: kernel function parameters and error penalty coefficients.

In this embodiment, the parameters of the obtained classification model can also be optimized, specifically: the genetic algorithm is used to optimize the kernel function parameters and error penalty coefficients in the support vector machine algorithm to improve the accuracy of screening and classification, so that the classification The recognition rate of the model can be close to 90%.

In order to better understand the screening method for cascaded utilization of lithium batteries described in the embodiments of the present invention, the screening method for cascaded utilization of lithium batteries described in the embodiments of the present invention will be described in detail, as shown in Figure 2 and Figure 3, the method described in the embodiments of the present invention The step-by-step utilization screening method of lithium batteries can specifically include the following steps:

Step 1. Obtain sample data (training sample data and test sample data) and sample labels (training sample labels and test sample labels):

1. Collect 200 18650-type waste lithium batteries from different manufacturers as sample lithium batteries, use optical scanning equipment to obtain CT images of each lithium battery, and obtain CT images of 200 lithium batteries as sample data.

2. Measure the capacity and internal resistance parameters of these 200 lithium batteries one by one, and record the lithium batteries with a capacity lower than 10% and an internal resistance higher than 200mΩ as lithium batteries that cannot be used in cascading, denoted by -1, and other records as cascading The lithium battery used, denoted by 1, gets the sample label.

Step 2, process the obtained CT image, and extract the feature value information:

1. CT image preprocessing:

Since the machine can only operate on the data, the scanned CT image cannot be directly used as the content recognized by the machine. Only some characteristic information of the CT image can be selected, which is called the characteristic value. Since the calculation of the eigenvalues needs to calculate the image pixel matrix, the obtained CT images may have different degrees of white background, which will greatly interfere with the calculation. Therefore, in the process of use, it is first necessary to remove interference, that is, to remove a large white background, that is, to cut off the white background of the CT image, so as to facilitate the calculation of eigenvalues.

The program flow for cropping the white background is as follows:

The CT image is stored in the form of a matrix. Each data in the matrix records the pixel of the point in the image. The gray value matrix of the CT image can be obtained through the matlab software, that is, the image is read, and the boundary of the CT image is found in the matlab software. Record the pixel position information at the boundary, determine the clipping boundary, and use the clipping function to complete.

2. Calculate the eigenvalues:

Lithium batteries that can be used in steps usually have a clearer internal structure than lithium batteries that cannot be used in steps, and contrast is the main parameter to measure the clarity of grayscale images. At present, there are different methods for defining the contrast of grayscale images, and the contrast defined by a variety of different methods can be selected as the feature value, as shown in Figure 3, the contrast used in this embodiment includes: per pixel contrast, Weber contrast, mean square Root contrast, Michelson contrast.

Next, the contrast ratio defined by different methods is explained:

(1) Contrast per pixel C _pp

Per-pixel contrast is defined as the cumulative intensity difference between the current pixel and a ring of neighboring pixels:

Wherein, i=1:m, j=1:n; m, n are the size of the CT image grayscale image matrix; I(i, j) is the grayscale value of the pixel in row i and column j.

exist or In the boundary case of , the calculation of I(x, y) is skipped.

The average C _PP of the whole image is

(2) Weber contrast C _w

Weber contrast is defined as:

Wherein, I _b is the intensity value of the background. In this embodiment, the background is white, so I _b is equal to 255. Since I _b is always greater than or equal to I(i,j), |I(i,j)-I _b | can be replaced by _Ib -I(i,j).

The Weber contrast of the entire image is:

(3) Michelson contrast C _m

Define Michelson contrast as:

where, I _max (i,j)=arg max _{x∈[i-1,i+1],y∈[j-1,j+1]} I(x,y)

I _min (x,y)=arg min _{x∈[i-1,i+1],y∈[j-1,j+1]} I(x,y)

The average Michelson contrast for the entire image is:

(4) Root mean square contrast is defined as:

in,

The procedure for calculating the eigenvalues is as follows:

Read the CT grayscale image matrix, write a contrast calculation program according to different contrast definitions, and obtain 4 eigenvalues for each image of the sample lithium battery according to the above 4 contrast definitions, and store them in the EXCEL table; and use 1, -1 Indicates the sample label, where 1 indicates a lithium battery that can be used in steps, and -1 indicates a lithium battery that cannot be used in steps. Store the sample label information of the sample data in the EXCEL table; in this way, a lithium battery can get a set of corresponding records : 4 contrast parameters and label information.

Step 3, use matlab to write the support vector machine algorithm program according to the support vector machine theory, use 150 of them as the training sample data and training sample labels for training the classification model, and read the corresponding contrast parameters and parameters of the 150 from the EXCEL table data label information, train the classification model, and obtain the trained classification model.

Step 4, use the remaining 50 test sample data and test sample labels as the test classification model, test the classification model obtained after training, and output the prediction labels of these 50 lithium batteries through the classification model obtained after training; The real labels of these 50 lithium batteries (ie: test sample labels), match the predicted labels with the real labels, and obtain the classification recognition rate according to the degree of matching; if, at this time, the labels of 40 images are correctly predicted, then the classification model The classification recognition rate is 80%.

Step 5, optimize the parameters in the support vector machine algorithm to improve the classification recognition effect.

The classification effect of the support vector machine algorithm mainly depends on the kernel function parameter gama, which is γ, and the error penalty coefficient cost, which is C. Finding the optimal parameters can greatly improve the effect of the classification model. The gama and cost parameters in the support vector machine are optimized by genetic algorithm. The program flow is as follows:

1) Initialize the parameters of the Support Vector Machine (SVM), there are two main parameters that affect the classification effect of the SVM, namely the kernel function parameter gamma, referred to as γ, and the error penalty coefficient Cost, referred to as C, to determine the genetic algorithm The initial group, the number of groups can be selected as 100, and the two parameters (γ and C) that need to be optimized by the support vector machine are binary coded to obtain the initial 100 groups, where the length of the binary code string depends on the range of the SVM parameters to be searched and precision to determine;

2) Set the parameters of the genetic algorithm, the initial algebra, the initial crossover probability, the initial mutation probability, the maximum genetic algebra, etc.;

3) Decode the initial group and send it to SVM for training and calculate the fitness of the individual. The fitness function adopts the correct rate of SVM classification;

4) Apply the optimal preservation strategy, and save the individual with the highest fitness (correct rate) before performing the genetic operation, so as to prevent the loss of the excellent gene due to the genetic operator operation, and record the index of the worst individual;

5) Carry out genetic operations on the above-mentioned initial population, specifically: the selection operator adopts the gambler selection method, the crossover operator adopts single-point crossover, a new population is generated through the genetic operator operation, and the fitness degree (correct rate) saved in 4) is used ) The highest individual replaces the new individual whose serial number is Index, and finally generates a new group, and then returns to 3) for training;

6) Check whether the termination condition of the algorithm is satisfied: since the classification accuracy (fitness) itself is the result to be searched, it is difficult to be used as the termination condition, but when the fitness of the optimal individual of several consecutive generations is close to equal, it is considered that the population can no longer evolve , the algorithm is terminated; or the set maximum genetic algebra is used as the termination condition of the algorithm, and when any one of the above two conditions is met, the algorithm is automatically terminated. The recognition rate of the optimized classification model can be close to 90%.

It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is a relationship between these entities or operations. any such actual relationship or order exists between them.

The above description is a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, these improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (8)

1. A lithium battery echelon utilization screening method is characterized by comprising the following steps:

acquiring training sample data and a training sample label; wherein, training sample data is the CT image of the sample lithium cell that is used for training, training sample label includes: label information corresponding to the CT image, wherein the label information is used for identifying whether a corresponding sample lithium battery for training can be utilized in a gradient manner;

training a classification model to be trained according to the obtained training sample data and training sample labels to obtain a trained classification model;

and acquiring a CT image of the lithium battery to be tested, inputting the acquired CT image of the lithium battery to be tested into the trained classification model, and outputting whether the lithium battery to be tested can be utilized in a gradient manner or not by the trained classification model.

2. The lithium battery echelon utilization screening method of claim 1, wherein the obtaining of the training sample label comprises:

measuring the capacity and the internal resistance of a sample lithium battery for training;

and classifying the sample lithium battery with the capacity lower than a first preset threshold and the internal resistance higher than a second preset threshold into the lithium battery which cannot be utilized in a gradient manner, and classifying the other sample lithium batteries into the lithium batteries which can be utilized in the gradient manner.

3. The lithium battery echelon utilization screening method of claim 1, wherein before training a classification model to be trained according to acquired training sample data and training sample labels to obtain a trained classification model, the method further comprises:

performing image processing on the acquired training sample data, and calculating a contrast value of the training sample data as a characteristic value;

and taking the characteristic value of the training sample data obtained by calculation as the input of the classification model to be determined.

4. The lithium battery gradient utilization screening method of claim 3, wherein the contrast value comprises: per-pixel contrast, weber contrast, root mean square contrast, michelson contrast.

5. The lithium battery echelon utilization screening method of claim 3, wherein the training of the classification model to be trained according to the obtained training sample data and training sample labels to obtain the trained classification model comprises:

taking the characteristic value of the training sample data obtained by calculation as the input of a classification model to be determined, and taking the obtained training sample label as the output of the classification model to be trained;

and training the classification model to be trained by utilizing a neural network training technology in supervised learning, and constructing the classification model based on the support vector machine algorithm.

6. The lithium battery echelon utilization screening method of claim 1, wherein the trained classification model is a classification decision plane;

the method comprises the following steps of inputting the obtained CT image of the lithium battery to be tested into a trained classification model, and outputting the category of the lithium battery to be tested by the trained classification model, wherein the category comprises the following steps:

and inputting the obtained CT image of the lithium battery to be tested into the trained classification model, and determining whether the lithium battery to be tested can be utilized in a gradient manner according to the side of the classification decision plane where the input characteristic value of the lithium battery to be tested is located.

7. The lithium battery echelon utilization screening method of claim 1, wherein after training a classification model to be trained according to the acquired training sample data and training sample labels to obtain the trained classification model, the method further comprises:

obtaining test sample data and a test sample label; wherein, the test sample data is a CT image of a sample lithium battery for testing, and the test sample label comprises: label information corresponding to the CT image, wherein the label information is used for identifying whether a corresponding sample lithium battery for testing can be utilized in a gradient manner;

calculating a contrast value of the test sample data as a characteristic value;

taking the characteristic value of the test sample data obtained by calculation as the input of the trained classification model;

and matching and verifying the prediction label output by the trained classification model and the acquired corresponding test sample label.

8. The lithium battery echelon utilization screening method of claim 5, wherein after training a classification model to be trained by a neural network training technique in machine learning and constructing a classification model based on a support vector machine algorithm, the method further comprises:

optimizing parameters in a support vector machine algorithm by adopting a genetic algorithm, wherein the optimized parameters comprise: kernel function parameters and error penalty coefficients.