Battery cell anomaly detection method
Through the restricted Boltzmann model and Gauss Bernoulli energy function, the multimodal distribution problem in the detection of K value abnormality of the battery cell is solved, the detection accuracy and robustness are improved, and the battery safety and production optimization are ensured.
Patent Information
- Application Number
- CN202410907137.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-08
- Publication Date
- 2024-12-13
AI Technical Summary
The existing methods of detecting abnormal cell K value of battery cells are difficult to adapt to the situation of different cell numbers and complex K value statistical distribution, which can easily cause false alarms or missed alarms, and cannot effectively detect abnormal cell cells in the pallet.
Using the restricted Boltzmann model, by constructing Gauss Bernoulli energy function and optimizing model parameters, the free energy value is used to determine whether the cell sample is abnormal, and a restricted control statistics are constructed and the control limit is set to determine the abnormal cell.
It improves the accuracy and robustness of abnormal data detection, saves costs, ensures battery safety, and provides a basis for cell production optimization, improving product pass rate and safety.
Smart Images
Figure CN119125892A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of batteries, and in particular relates to a method for detecting abnormality of a battery cell. Background Art
[0002] Before the battery cells are shipped, they must be tested for K value, and the cells with abnormal K value (equivalent to self-discharge) must be picked out. The K value of the battery cell is one of the important indicators for evaluating the health status and performance of the battery, and has a direct impact on the quality and safety of the battery cell. The existing abnormal K value detection of battery cells is mainly carried out on a pallet basis, which requires centralized analysis and determination of the K values of multiple cells in each pallet. However, in actual applications, the number of cells in each pallet is significantly different, ranging from a few to dozens, and the statistical distribution of the K value also presents a high degree of complexity. Traditional detection methods are difficult to adapt to such a situation with different numbers of cells and complex distribution, and are prone to false alarms or missed alarms. Therefore, there is an urgent need for a new K value anomaly detection method that can handle pallets with different numbers of cells and adapt to their complex K value statistical distribution to improve the accuracy and robustness of detection. Summary of the invention
[0003] In order to solve the above technical problems, the present invention provides a method for detecting abnormality of a battery cell.
[0004] The present invention is achieved through the following technical solutions: A method for detecting abnormality of a battery cell comprises the following steps: Obtain the cell data of a specified batch and perform preprocessing to form training samples; Constructing restricted Boltzmann models; Optimizing the parameters of the restricted Boltzmann model through the training samples; Constructing and solving a restricted control metric for the restricted Boltzmann model; When the free energy value of the battery cell sample exceeds the limited control statistic range, the battery cell sample is determined to be abnormal.
[0005] In some embodiments, the battery cell data includes: battery cell time, batch number, K value, DC resistance, AC resistance, case pressure and internal resistance data.
[0006] In some embodiments, the preprocessing includes constructing a battery cell data list according to the time of the battery cell and the batch number; The battery cell data list includes K value, DC resistance, AC resistance, case pressure and internal resistance value; Performing noise removal on the cell data list; The cell data list after denoising is normalized.
[0007] In some embodiments, constructing the restricted Boltzmann model includes constructing a Gaussian Bernoulli energy function based on a given state; A joint probability distribution of the states is obtained based on the Gaussian Bernoulli energy function.
[0008] In some embodiments, the parameters of the restricted Boltzmann model are optimized, and the likelihood function is logarithmized to obtain a log-likelihood function; Maximizing the log-likelihood function by a gradient ascent method; Based on the training samples, K-step Gibbs sampling is performed using a contrast divergence algorithm to obtain parameters of an optimized restricted Boltzmann model.
[0009] In some embodiments, based on the battery cell samples, a free energy function is set as a restricted control statistic.
[0010] In some embodiments, the restricted control statistic is solved by analyzing the free energy function of the battery cell sample to obtain the mean and standard deviation of the restricted control statistic.
[0011] In some embodiments, the free energy value is obtained by a free energy function.
[0012] In some embodiments, the restricted control statistic range includes an upper control limit and a lower control limit.
[0013] In some embodiments, the upper control limit and the lower control limit are obtained by the mean and standard deviation of the restricted control statistic.
[0014] Compared with the prior art, the present invention has the following advantages: 1. Improved the ability to detect abnormal data. By constructing a Gauss-Boltzmann model, it can adapt to the multimodal distribution of data and determine whether the sample data is abnormal by the free energy value; 2. Cost saving: since faulty cells can be detected in advance, ineffective subsequent processing can be avoided, thus saving material and energy costs. At the same time, the qualified rate of the entire output is also greatly improved due to the detection and elimination of abnormal cells; 3. Improve battery safety. If abnormal cells cannot be detected, safety accidents may occur during battery assembly and use. The present invention can effectively detect abnormal cells and ensure the safety of battery use. 4. Provide a basis for optimizing the production of battery cells. The test results can guide the adjustment and optimization of process flow and equipment parameters to reduce the generation of abnormal battery cells.
[0015] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the present use information. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0017] Figure 1 The flowchart of an embodiment is shown in FIG. DETAILED DESCRIPTION
[0018] The drawings are only for illustrative purposes and cannot be construed as limiting the present invention. To better illustrate the present embodiment, some parts of the drawings may be omitted, enlarged, or reduced, and do not represent the size of the actual product. For those skilled in the art, it is understandable that some well-known structures and their descriptions may be omitted in the drawings. The positional relationships described in the drawings are only for illustrative purposes and cannot be construed as limiting the present invention.
[0019] Figure 1 The present invention provides a flow chart of a method for detecting abnormality of a battery cell, which includes the following steps: S110, pre-processing battery cell data.
[0020] Specifically, obtain the cell data of a specified batch, filter out the time, batch number, K value, DC resistance, AC resistance, case pressure and internal resistance data of all cells in the selected batch in the capacity division stage, and list the cell data of several cells of the same time and batch. ,in Represents the K value, DC resistance, AC resistance, case pressure, and internal resistance of the Mth battery cell.
[0021] The collected battery cell data is processed to remove noise, and the batteries with values higher than the fixed threshold K in each tray are removed. The K values of all the remaining batteries are used as the detection data. Therefore, the remaining battery cell data list is ,in Represents the cell K value, DC resistance, AC resistance, case pressure, internal resistance value of the nth cell and .
[0022] In addition, the denoised cell data list is mapped to the interval [-1, 1] through linear function normalization, effectively avoiding the influence of different channel feature weights caused by different dimensions. The specific formula is: ; (1) in, represents the normalized K value of the jth battery cell, 0 <j n, Represents the cell data list after denoising The jth element in and Represent the maximum and minimum values in each characteristic sequence respectively. The normalized cell K value list is as follows: , and use it as a training sample.
[0023] S120. Construct a restricted Boltzmann model.
[0024] Specifically, we first construct the Gaussian Bernoulli energy function. ; (2) Among them, given a set of random vectors v and h forming the state (v, h), v represents the state of the visible layer, is the state of the i-th visible variable, h is the hidden layer state, is the state of the jth hidden variable, is the bias of the visible variable, is the bias of the hidden variable, is the connection weight between the visible variable and the hidden variable, visible is the number of nodes in the visible layer, hidden is the number of nodes in the hidden layer, 0 visible,0 <j hidden.
[0025] According to the Gaussian Bernoulli energy function, the joint probability of the state (v, h) is ; (3) in, Represents the set of parameters for all connection weights and biases, including, , and , is the bias of the visible variable, is the bias of the hidden variable, is the connection weight between visible and hidden variables, and is the partition function formula, specifically, ; (4) Furthermore, the values of the visible and hidden layers of the restricted Boltzmann machine usually follow Binary distribution, in the continuous distribution scenario in the industrial field, by improving the model of Gauss-Bernoulli energy function (2), the Gauss-Bernoulli Boltzmann model is used, and its energy function is defined as: ; (5) Among them, the status The x is the visible layer state, and the y is the hidden layer state. is the state of the ith visible variable, is the state of the jth hidden variable, is the bias of the visible variable, is the bias of the hidden variable, is the connection weight between visible variables and hidden variables, is the standard deviation of the visible layer variable, each visible variable obey( , ), visible is the number of visible layer nodes, hidden is the number of hidden layer nodes, 0 visible,0 <j hidden.
[0026] Usually for the joint probability distribution The hidden variables in cannot be solved, so we can observe the probability distribution of the visible layer data To infer the Boltzmann function, sum the hidden layer state vectors and get the formula as follows: ; (6) S130. Optimize parameters of the restricted Boltzmann model.
[0027] Specifically, the K value training data of the battery cell is obtained by using S110 as a training sample, and the parameters are adjusted according to the maximum likelihood criterion. , in order to fit the given training samples, so that the probability distribution represented by the restricted Boltzmann model under the parameters is as consistent as possible with the training samples.
[0028] Assume that the training sample X=( ), n is the training sample size, each data satisfies independent and identical distribution, and the training goal is to maximize the likelihood function , log the likelihood function and simplify the calculation to get the formula of log likelihood function as follows: ; (7) Among them, the parameters Represents the set of parameters for all connection weights and biases, including, , and , is the bias of the visible variable, is the bias of the hidden variable, is the connection weight between the visible variable and the hidden variable.
[0029] Furthermore, the gradient ascent method is used to maximize the log-likelihood function. The iterative formula is as follows: ; (8) in , is the learning rate.
[0030] Furthermore, the contrastive divergence (CD) algorithm is used to learn the model parameters, and the CD algorithm is used to perform K Gibbs sampling. The specific steps are as follows: (1) Yes , take the initial value , after k steps of Gibbs sampling, the samples are generated , the sampling formula is as follows: ; (9) ; (10) Among them, X is the training sample, x is the visible layer state, y is the hidden layer state, is the state of the ith visible variable, is the state of the jth hidden variable, is the bias of the visible variable, is the bias of the hidden variable, is the connection weight between visible variables and hidden variables, is the sigmoid activation function.
[0031] (2) Execute the Sth step (s=1,2,...k) successively, from Sampling out ,from Sampling out , using K-step Gibbs sampling to obtain The approximate gradient is calculated as follows: ; (11) ; (12) ; (13) in is the learning rate, is the positive gradient, is the reverse gradient, is the bias of the updated visible variable, is the bias of the updated hidden variable, is the connection weight between the visible variable and the hidden variable.
[0032] (3)Set k to 1, that is, only perform Gibbs sampling once.
[0033] S140. Construct a restricted control statistic and solve it.
[0034] Specifically, in the Gaussian Boltzmann model, each visible layer neuron follows a Gaussian distribution. The marginal distribution of the visible layer neurons refers to the probability density function (PDF) of each visible layer neuron. These marginal distributions describe the Gaussian degree of each neuron, that is, the probability density of the values that the neuron can take. Given the values of the visible layer neurons, the joint probability density of all visible layer neurons. The higher the joint probability density, the more the distribution of the data in the visible layer space conforms to the characteristics of the Gaussian distribution, the higher the probability of each neuron, and the more likely the data is normal. On the contrary, the lower the joint probability density, the less the distribution of the data in the visible layer space conforms to the characteristics of the Gaussian distribution, the lower the probability of each neuron, and the more likely the data is abnormal. Therefore, the degree of abnormality can be inversely proportional to the joint probability density of the data, that is, the lower the joint probability density, the more likely the data is abnormal.
[0035] According to the formula: ; (14) It can be seen that the free energy function F(x) has a negative linear relationship with the marginal distribution of the visible layer data, and the marginal distribution of the visible layer neurons also follows a standard Gaussian distribution, that is, the free energy function F(x) follows a Gaussian distribution , therefore, the free energy function is used as the restricted control statistic as the basis for fault detection.
[0036] Furthermore, perform statistical analysis on the free energy values of normal sample data, and calculate the mean and standard deviation . Based on the statistical analysis results, set the upper control limit (UCL) to , and the lower control limit (LCL) to .
[0037] S150. Detect anomalies in the sample based on the trained model.
[0038] Specifically, for new sample data, calculate the free energy value of its free energy function F(x). If F(x) exceeds the control limit range (i.e., F(x) < UCL or F(x) > LCL), then the sample is determined to be abnormal.
[0039] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for detecting abnormality of a battery cell, characterized in that: include: Obtain the cell data of a specified batch and perform preprocessing to form training samples; Constructing restricted Boltzmann models; Optimizing the parameters of the restricted Boltzmann model through the training samples; Constructing and solving a restricted control metric for the restricted Boltzmann model; When the free energy value of the battery cell sample exceeds the limited control statistic range, the battery cell sample is determined to be abnormal.
2. A method for detecting abnormality of a battery cell according to claim 1, characterized in that: The battery cell data includes: battery cell time, batch number, K value, DC resistance, AC resistance, case pressure and internal resistance data.
3. A method for detecting abnormality of a battery cell according to claim 2, characterized in that: The preprocessing includes constructing a battery cell data list according to the time of the battery cell and the batch number; The battery cell data list includes K value, DC resistance, AC resistance, case pressure and internal resistance value; Performing noise removal on the cell data list; The cell data list after denoising is normalized.
4. A method for detecting abnormality of a battery cell according to claim 1, characterized in that: Constructing the restricted Boltzmann model includes constructing a Gaussian Bernoulli energy function based on a given state; A joint probability distribution of the states is obtained based on the Gaussian Bernoulli energy function.
5. A method for detecting abnormality of a battery cell according to claim 1, characterized in that: Optimizing the parameters of the restricted Boltzmann model and logarithmizing the likelihood function to obtain a log-likelihood function; Maximizing the log-likelihood function by a gradient ascent method; Based on the training samples, K-step Gibbs sampling is performed using a contrast divergence algorithm to obtain parameters of an optimized restricted Boltzmann model.
6. A method for detecting abnormality of a battery cell according to claim 1, characterized in that: Based on the cell samples, a free energy function is set as a restricted control statistic.
7. A method for detecting abnormality of a battery cell according to claim 1 or claim 6, characterized in that: The restricted control statistic is solved, and the mean and standard deviation of the restricted control statistic are obtained by analyzing the free energy function of the battery cell sample.
8. A method for detecting abnormality of a battery cell according to claim 6, characterized in that: The free energy value is obtained by a free energy function.
9. A method for detecting abnormality of a battery cell according to claim 6, characterized in that: The restricted control statistic range includes an upper control limit and a lower control limit.
10. A method for detecting abnormality of a battery cell according to claim 9, characterized in that: The upper control limit and the lower control limit are obtained by the mean and standard deviation of the restricted control statistic.