Abnormal battery cell detection method, device, electronic device and storage medium

Abstract

The embodiments of the present application provide a method, device, electronic device, and computer-readable storage medium for detecting abnormal battery cells, and relate to the field of artificial intelligence. The method includes: obtaining test information of multiple battery cells to be tested, each test information being a parameter set of the corresponding battery cell to be tested under multiple processes, and each parameter set including at least one test parameter of the corresponding process; inputting the test data of the multiple battery cells to be tested into a pre-trained state prediction model to obtain a state sequence of the multiple battery cells to be tested output by the state prediction model; inputting the state sequence of the multiple battery cells to be tested into a pre-trained anomaly detection model according to the production order of the battery cells to be tested, and obtaining the detection results of the multiple battery cells to be tested output by the anomaly detection model. The embodiments of the present application improve detection efficiency and detection accuracy.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology. Specifically, the present application relates to a method, device, electronic device, computer-readable storage medium and computer program product for detecting abnormal battery cells. Background Art

[0002] In the battery manufacturing process, short-circuit testing (Hipot testing) is a critical step in the battery production line to detect abnormal particles, burrs, and separator damage. Short-circuit testing is primarily performed during the winding, hot pressing, pre-welding, butterfly welding, shell insertion, and primary injection processes.

[0003] Existing short-circuit tests mainly apply high voltage to the battery. If the battery's insulation is good enough, the large voltage difference between the positive and negative poles of the battery will only cause a small current to flow. Although this tiny current is acceptable, the battery itself should not break down because the short-circuit test is always a non-destructive test.

[0004] Because the short-circuit test is a non-destructive test, the parameter settings are often conservative. Among the batteries that pass the drop voltage test, there are often some that fail the actual short-circuit test, resulting in low detection accuracy. Summary of the Invention

[0005] The embodiments of the present application provide a method, device, electronic device, computer-readable storage medium, and computer program product for detecting abnormal battery cells, which can solve the above-mentioned problems of the prior art. The technical solution is as follows:

[0006] According to one aspect of an embodiment of the present application, a method for detecting abnormal battery cells is provided, the method comprising:

[0007] Acquire test information of multiple battery cells to be tested, where each piece of test information is a parameter set of the corresponding battery cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process;

[0008] Inputting test data of multiple cells to be tested into a pre-trained state prediction model to obtain a state sequence of multiple cells to be tested output by the state prediction model, wherein the state sequence includes a first latent state of the cell to be tested in each process, and each first latent state is used to indicate whether the corresponding cell to be tested is abnormal in the corresponding process;

[0009] Inputting the state sequences of the plurality of cells to be tested into a pre-trained anomaly detection model according to the production order of the cells to be tested, and obtaining detection results of the plurality of cells to be tested output by the anomaly detection model, wherein the detection results are used to indicate whether each cell to be tested is abnormal;

[0010] The state prediction model is formed by training a Hidden Markov Model (HMM) model using the test information of the sample battery cell as a training sample and the state sequence of the sample battery cell as a training label;

[0011] The anomaly detection model is formed by training a correlation gated recurrent (CGRU) model using a state sequence of multiple sample cells arranged in the production order of the sample cells as training samples and a detection result of multiple sample cells arranged in the production order of the sample cells as training labels.

[0012] According to another aspect of an embodiment of the present application, a device for detecting abnormal battery cells is provided, the device comprising:

[0013] A test information acquisition module is used to obtain test information of multiple cells to be tested, each of which is a parameter set of the corresponding cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process;

[0014] A state sequence prediction module is used to input the test data of multiple cells to be tested into a pre-trained state prediction model to obtain a state sequence of the multiple cells to be tested output by the state prediction model, wherein the state sequence includes the first latent state of the cells to be tested in each process, and each first latent state is used to indicate whether the corresponding cell to be tested is abnormal in the corresponding process;

[0015] a detection model output module, configured to input the state sequences of the plurality of cells to be tested into a pre-trained anomaly detection model according to the production order of the cells to be tested, and obtain detection results of the plurality of cells to be tested output by the anomaly detection model, wherein the detection results are used to indicate whether each cell to be tested is abnormal;

[0016] The state prediction model is formed by training a Hidden Markov Model (HMM) model using the test information of the sample battery cell as a training sample and the state sequence of the sample battery cell as a training label;

[0017] The anomaly detection model is formed by training a correlation gated recurrent (CGRU) model using a state sequence of multiple sample cells arranged in the production order of the sample cells as training samples and a detection result of multiple sample cells arranged in the production order of the sample cells as training labels.

[0018] According to another aspect of an embodiment of the present application, an electronic device is provided. The electronic device includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the above method.

[0019] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0020] According to one aspect of an embodiment of the present application, a computer program product is provided, including a computer program, which implements the steps of the above method when executed by a processor.

[0021] The beneficial effects of the technical solution provided by the embodiments of the present application are:

[0022] First, test information of multiple battery cells to be tested is obtained, where each test information is a parameter set of the corresponding battery cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process. The test data of the multiple battery cells to be tested are input into a pre-trained state prediction model to obtain state sequences of the multiple battery cells to be tested output by the state prediction model. The state sequences of the multiple battery cells to be tested are then used as inputs to an anomaly detection model. The anomaly detection model learns the autocorrelation within the time series constituted by the state sequences of the multiple battery cells to be tested and the correlation between different features, and finally obtains the detection results to perform anomaly screening on the battery cells, thereby improving detection efficiency and detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments of the present application.

[0024] Figure 1 A schematic flow chart of a method for detecting abnormal battery cells provided in an embodiment of the present application;

[0025] Figure 2 A schematic diagram of a training process of a state prediction model provided in an embodiment of the present application;

[0026] Figure 3 A schematic diagram of the training process of an anomaly detection model provided in an embodiment of the present application;

[0027] Figure 4 A schematic diagram of the process flow of each iteration of a CGRU model provided in an embodiment of the present application;

[0028] Figure 5 A schematic structural diagram of a device for detecting abnormal battery cells provided in an embodiment of the present application;

[0029] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] The following describes the embodiments of the present application in conjunction with the accompanying drawings. It should be understood that the embodiments described below in conjunction with the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0031] Those skilled in the art will understand that, unless otherwise stated, the singular forms "a", "an" and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application mean that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the present technical field. It should be understood that when we say that an element is "connected" or "coupled" to another element, the element can be directly connected or coupled to the other element, or it can refer to that the element and the other element establish a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used here can include wireless connection or wireless coupling. The term "and / or" used here indicates at least one of the items defined by the term, for example, "A and / or B" can be implemented as "A", or as "B", or as "A and B".

[0032] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0033] In the battery process, short-circuit testing (Hipot testing) and K-value testing are important processes in the battery production line to detect abnormal particles, burrs, and diaphragm damage in the battery. Short-circuit testing is primarily performed during the winding, hot pressing, pre-soldering, butterfly welding, shell insertion, and primary liquid filling processes. K-value testing is performed after OCV3 and OCV4 (OCV (Open Circuit Voltage) testing) measures the potential difference between the two electrodes of a battery in the open-circuit state without discharge; OCV testing is important for assessing battery health, state of charge (SOC), and predicting battery performance). Prior to this, the battery process also includes resting, forming, resting, secondary liquid filling, cleaning, volume separation, and constant-temperature resting. This is not only time-consuming but also tests for different types of foreign matter. Therefore, if the battery can be tested for anomalies again based on the results of six short-circuit tests to screen out problematic cells, significant time and material costs can be saved. Furthermore, implementing server deployment and enabling the production line's MES to call an interface for robotic automatic grabbing and intelligent sorting and waste removal can significantly reduce labor costs.

[0034] Existing battery short-circuit testing technology mainly applies high voltage to the battery. If the battery's insulation is good enough, the large voltage difference between the positive and negative poles of the battery will only cause a small current to flow. Although this tiny current is acceptable, the battery itself should not break down because the short-circuit test is always a non-destructive test.

[0035] The main problems of the existing technology are:

[0036] 1. The difference between DC voltage and AC voltage

[0037] An AC hipot (alternating current short-circuit) test does not require discharging the DUT (device under test) after the test, and voltage is applied alternately at both poles of the DUT, resulting in more stringent test results. A DC hipot (direct current short-circuit) test, on the other hand, uses a lower current level and requires a higher test voltage. Voltage is applied only at one pole, and the DUT must be discharged after the test, which is more time-consuming. Therefore, an AC hipot test is generally used for battery short-circuit testing. However, the AC hipot test only measures the total leakage current and cannot properly detect transient discharge currents caused by factors such as particulate matter and glitches. Therefore, the voltage drop after the voltage is applied is often used as the test result.

[0038] 2. The short-circuit test is a non-destructive test, so the parameter setting is relatively conservative. Among the cells that pass the drop voltage test, there are often some that fail the actual short-circuit test. Considering the need to save materials, time and manpower, it is very necessary to conduct another short-circuit abnormality detection on the cell based on the data of the first six short-circuit tests before filling the liquid, and the impact of the six short-circuit test results must be reflected.

[0039] 3. The K value can effectively detect foreign matter between the positive electrode and the diaphragm. The short-circuit test can effectively detect the complete penetration of the positive and negative electrodes caused by foreign matter. However, it is difficult to detect foreign matter between the negative electrode and the diaphragm.

[0040] The abnormal battery cell detection method, device, electronic device, computer-readable storage medium and computer program product provided in the present application are intended to solve the above technical problems of the prior art. By training a Hidden Markov Model (HMM) model, the model will be able to identify hidden states in an input time series; and use the predict method of the HMM model to assign the most likely hidden state (also referred to as the first hidden state) to each process; then use the hidden state identified by the HMM model as the input feature of the Correlation Gated Recurrent Unit (CGRU) model to predict the predicted value for a given hidden state, and then detect short-circuited abnormal battery cells through the output (observation value). The embodiments of the present application can improve the detection rate of abnormal battery cells with foreign matter between the negative electrode and the diaphragm of the battery cell.

[0041] The following describes several exemplary embodiments to illustrate the technical solutions of the embodiments of the present application and the technical effects produced by the technical solutions of the present application. It should be noted that the following embodiments can refer to, draw on, or combine with each other, and the same terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0042] The present application provides a method for detecting abnormal cells. Figure 1 As shown, the method includes:

[0043] S101: Acquire test information of multiple battery cells to be tested.

[0044] The test information of each battery cell to be tested in the embodiment of the present application is a parameter set of the corresponding battery cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process.

[0045] As can be seen from the above description, this application uses multiple processes to test each battery cell to be tested, and collects test parameters for each process. In some embodiments, the multiple processes of the embodiments of the present application can include: coil cutting short circuit test, hot pressing short circuit test, pre-welding short circuit test, butterfly welding short circuit test, shell entry short circuit test, and single liquid injection short circuit test. By performing short circuit tests on the battery cell to be tested from six aspects, a rich set of test parameters can be obtained, which is conducive to improving the accuracy of detection.

[0046] Please refer to Table 1, which exemplarily shows the information table of each test parameter in the parameter set of the embodiment of the present application. It can be seen that the parameter machine of the embodiment of the present application contains information of 12 test parameters, and provides the conditions for determining battery cell abnormality for some test parameters. For example, this parameter is the voltage drop parameter during the continuous charging phase. If the actual drop value is greater than the set threshold value, the test item fails.

[0047]

[0048] Table 1 Test parameter information table

[0049] The test information Z(i) of the battery cell i to be tested in the embodiment of the present application can be expressed by the following formula 1:

[0050] Formula 1

[0051] Among them, z s Represents the parameter set of the cell i under process s. Taking Table 1 as an example, each parameter set can be expressed as .

[0052] S102 , inputting test data of the plurality of battery cells to be tested into a pre-trained state prediction model to obtain a state sequence of the plurality of battery cells to be tested output by the state prediction model.

[0053] The state prediction model of the embodiment of the present application is obtained by training the HMM model. Specifically, the state prediction model is trained by using the test information of the sample battery cell as a training sample and the state sequence of the sample battery cell as a training label, so that the state prediction model can find the optimal possible state sequence given a test information.

[0054] The state sequence of the embodiment of the present application includes the first hidden state of the corresponding battery cell (battery cell to be tested or sample battery cell) in each process. Each first hidden state is used to characterize whether the corresponding battery cell to be tested is abnormal in the corresponding process. For example, the state sequence of a certain battery cell is [0,0,0,1,0,1]. The state sequence has 6 elements, indicating whether the battery cell is abnormal in the 6 processes. 0 indicates abnormal and 1 indicates normal. It should be noted that this application can also use other characters or numbers to indicate normal or abnormal, which will not be repeated here.

[0055] S103 , inputting the state sequences of the plurality of cells to be tested into a pre-trained anomaly detection model according to the production order of the cells to be tested, and obtaining detection results of the plurality of cells to be tested output by the anomaly detection model.

[0056] It is important to note that this application uses a batch detection approach when performing abnormal cell detection to further improve detection efficiency. The anomaly detection model is derived from the CGRU model, which can perform time series decomposition within neurons and learn the correlation (autocorrelation and cross-correlation) between each feature and the relationship between states.

[0057] Specifically, the anomaly detection model is trained on the CGRU model using the state sequences of multiple sample cells arranged in the production order of the sample cells as training samples and the detection results of multiple sample cells arranged in the production order of the sample cells as training labels.

[0058] In an embodiment of the present application, the state sequences of multiple sample cells are input into the CGRU model according to the production order of these sample cells, so that the CGRU model can recursively update the predicted detection results of each sample cell according to the production order of the sample cells.

[0059] In an embodiment of the present application, the state sequences of the multiple battery cells to be tested are input into a pre-trained anomaly detection model according to the production order of the battery cells to be tested, and the detection results of the multiple battery cells to be tested output by the anomaly detection model are obtained by performing time series analysis and correlation analysis.

[0060] The abnormal battery cell detection method of the embodiment of the present application first obtains test information of multiple battery cells to be tested, each test information is a parameter set of the corresponding battery cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process. The test data of the multiple battery cells to be tested are input into a pre-trained state prediction model to obtain the state sequence of the multiple battery cells to be tested output by the state prediction model, and the state sequence of each of the multiple battery cells to be tested is continued to be used as the input of the abnormality detection model. The abnormality detection model learns the autocorrelation within the time series constituted by the state sequence of the multiple battery cells to be tested and the correlation between different features, and finally obtains the detection result to perform abnormal screening of the battery cells, thereby improving the detection efficiency and detection accuracy.

[0061] On the basis of the above embodiments, considering that the battery cells generate a large number of initial detection parameters in each process, in order to further improve the detection efficiency, the embodiments of the present application can screen the initial detection parameters generated in each process before training the model to obtain one or more most representative detection parameters. Specifically, the embodiments of the present application can obtain the initial test information and test results of multiple sample battery cells. The initial test information includes multiple initial test parameters. Through a preset scoring function, the significance relationship between each initial test parameter and the test result is analyzed. The higher the significance, the higher the evaluation score. At least one initial test parameter with the highest evaluation score is used to construct the test information for model training and inference.

[0062] Specifically, the embodiment of the present application can create a matrix , The dimension is , is the number of cells, 12 is the 12 initial test parameters; Represents the iInitial test parameters corresponding to each battery cell j The frequency of The screening logic of the initial test parameter frequency in the embodiment of the present application is to count only valid data, so the count ratio matrix is ​​constructed as follows:

[0063]

[0064] Furthermore, the battery i The accuracy of each initial test parameter can be expressed by the following formula:

[0065]

[0066] Further battery cells i The information entropy of each initial test parameter can be expressed by the following formula:

[0067]

[0068] battery cells i The weight of each initial test parameter can be expressed by the following formula:

[0069]

[0070] Therefore, the score function of the embodiment of the present application can be expressed as:

[0071]

[0072] Based on the above embodiments, as an optional embodiment, see Figure 2 , the state prediction model is trained in the following way:

[0073] S201, initializing the initial state probability matrix, state transition probability matrix and emission state probability matrix of the HMM model;

[0074] S202: Construct an observation sequence, where the observation sequence is test information of the sample battery cell;

[0075] S203: Training and optimizing the parameters of the HMM model using the observation sequence, obtaining the corresponding relationship between the observation sequence and the training labels and the rule of mutual transition between the first hidden states of adjacent processes, and outputting the optimal state transition probability matrix and emission state probability matrix;

[0076] S204 : Obtain an optimal HMM model based on the optimal state transition probability matrix and the emission state probability matrix as the state prediction model.

[0077] It should be noted that, in the embodiment of the present application, the observation sequence of a sample cell is first defined by the above formula 1. The first hidden state can be expressed as {0, 1}. The most likely first hidden state of the cell can be expressed by the following formula 2:

[0078] Formula 2

[0079] The HMM model consists of the initial state probability matrix π, the state transition probability matrix A and the emission state probability matrix B.

[0080] By calculating the weight of each Gaussian probability density function , mean and covariance matrix , the continuous probability density function can be expressed as:

[0081] Formula 3

[0082] Given the first hidden state , observation sequence The probability distribution of can be expressed as a weighted sum of multiple Gaussian distributions, M is the number of individual Gaussian distributions, It is k Gaussian distribution weights, It is k The mean of a Gaussian distribution, It is k The covariance matrix of a Gaussian distribution.

[0083] and According to the observation sequence , is iterated by the Expectation-Maximization algorithm (EM), where the first hidden is calculated by the Viterbi algorithm.

[0084] It's important to note that time series prediction is a traditional problem in the field of artificial intelligence. Models such as recursive neural networks (RNNs), long short-term memory (LSTMs), and GRUs help improve time series prediction accuracy. However, because these methods learn independent models for each feature, they are unable to learn the relationships between time series features or between states.

[0085] The embodiment of the present application adopts the CGRU model, that is, based on the GRU model, further adds correlation analysis, specifically, Figure 3As shown, the anomaly detection model of the embodiment of the present application is trained in the following manner:

[0086] S301: Input the state sequences of multiple sample cells arranged in the production order of the sample cells into the CGRU model for time series analysis and correlation analysis to obtain the residual features of the state sequence of each sample cell, the second latent state of the seasonal features, and the second latent state of the trend features;

[0087] S302, sequentially obtaining prediction detection results for each sample battery cell based on the production order of the sample battery cells and the residual features of the state sequence of each sample battery cell, the second latent state of the seasonal feature, and the second latent state of the trend feature;

[0088] S303 : Based on the predicted detection results of each sample battery cell and the differences between the detection results, the CGRU model is trained and its parameters are optimized to obtain the optimal CGRU model (i.e., the one that meets the training end condition) as the anomaly detection model.

[0089] It should be noted that the time series of the embodiment of the present application refers to the state sequence of multiple sample battery cells arranged in the production order of the sample battery cells. That is to say, when the embodiment of the present application believes that there is an abnormality in the continuously generated battery cells, there may also be abnormalities in the battery cells generated before and after them.

[0090] Residuals, seasonality, and trend are three important concepts in time series analysis that together constitute the basic components of a time series. Seasonality refers to recurring cyclical fluctuations in a time series. A trend refers to a sustained upward or downward movement in a time series over a long period of time. Trend characteristics reflect the overall direction of the time series and may be linear (such as steady growth or decline) or nonlinear (such as accelerating growth or decelerating decline). Residuals, also known as irregular fluctuations or the remainder, refer to the portion of time series data that remains after removing trend and seasonality. Residuals are often considered random fluctuations or noise, which can be caused by a variety of factors, such as temporary events, measurement errors, or random variations within the system. Residuals are important in time series analysis because they provide important information about the stability of the time series and the accuracy of forecasts.

[0091] In time series analysis, residuals, seasons, and trends are interrelated. By decomposing a time series into these three components, embodiments of the present application can provide a deeper understanding of the structure and characteristics of the time series, thereby developing more accurate forecasting models and optimization strategies. Specifically, various time series decomposition methods (such as STL decomposition and seasonal decomposition) can be used to decompose the time series into three characteristics: trend, season, and residuals, and analyze and predict them separately.

[0092] Based on the above embodiments, as an optional embodiment, the state sequences of multiple sample cells arranged in the production order of the sample cells are input into the CGRU model for time series analysis and correlation analysis to obtain the residual features, the second latent state of the seasonal features, and the second latent state of the trend features of each sample cell, including:

[0093] The state sequences of multiple sample cells arranged in the production order of the sample cells are input into the CGRU model for multiple rounds of iterations;

[0094] For each round of iteration, Figure 4 Shown, including:

[0095] S401 , performing time series analysis on the state sequence of each sample battery cell according to the production order of the sample battery cells, and obtaining the residual features, seasonal features, and trend features of the state sequence of each sample battery cell in this round of iteration.

[0096] S402a. For each sample battery cell, based on the seasonal characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration, obtain a first correlation degree between the seasonal characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration, as the first correlation degree corresponding to the sample battery cell.

[0097] It should be noted that, for the first sample battery cell, since it does not have a previous sample battery cell, when calculating the first correlation degree corresponding to the first sample battery cell, the first preset value can be used as the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in this round of iteration, thereby obtaining the first correlation degree of the first sample battery cell based on the seasonal characteristics of the state sequence of the first sample battery cell in this round of iteration and the first preset value.

[0098] S402b. For each sample battery cell, based on the trend characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the trend characteristics of the state sequence of the previous sample battery cell in the current iteration, obtain a second correlation degree between the trend characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration, as the second correlation degree corresponding to the sample battery cell.

[0099] It should be noted that, for the first sample battery cell, since it does not have a previous sample battery cell, when calculating the second correlation degree corresponding to the first sample battery cell, the second preset value can be used as the second hidden state of the trend characteristics of the state sequence of the previous sample battery cell in this round of iteration, thereby obtaining the second correlation degree of the first sample battery cell based on the trend characteristics of the state sequence of the first sample battery cell in this round of iteration and the second preset value.

[0100] S403a. For each sample cell, obtain the second hidden state of the seasonal characteristics of the sample cell in the current iteration according to the first correlation degree corresponding to the sample cell, the seasonal characteristics in the current iteration, and the second hidden state of the seasonal characteristics of the previous sample cell in the current iteration.

[0101] It should be noted that after obtaining the first correlation degree corresponding to the sample battery cell, the seasonal characteristics of the sample battery cell in this round of iteration and the hidden state of the seasonal characteristics of the previous sample battery cell in this round of iteration can be further combined to obtain the second hidden state of the seasonal characteristics of the sample battery cell in this round of iteration. Similarly, since the first sample battery cell does not have a previous sample battery cell, the second hidden state of the seasonal characteristics of the sample battery cell in this round of iteration can be obtained based on the first correlation degree corresponding to the sample battery cell, the seasonal characteristics of this round of iteration and the first preset value.

[0102] S403b. For each sample cell, obtain the second hidden state of the trend feature of the sample cell in the current iteration according to the second correlation degree corresponding to the sample cell, the trend feature in the current iteration, and the second hidden state of the trend feature of the previous sample cell in the current iteration.

[0103] It should be noted that after obtaining the second correlation degree corresponding to the sample battery cell, the trend characteristics of the sample battery cell in this round of iteration and the second hidden state of the trend characteristics of the previous sample battery cell in this round of iteration can be further combined to obtain the second hidden state of the seasonal trend characteristics of the sample battery cell in this round of iteration. Similarly, since the first sample battery cell does not have a previous sample battery cell, the second hidden state of the trend characteristics of the sample battery cell in this round of iteration can be obtained based on the first correlation degree corresponding to the sample battery cell, the trend characteristics of this round of iteration and the second preset value.

[0104] Based on the above embodiments, as an optional embodiment, the first correlation degree of each sample cell in the embodiment of the present application includes the correlation and autocorrelation between the state sequence of the sample cell and the second hidden state of the seasonal characteristics of the previous sample cell in the current iteration;

[0105] The second correlation degree of each sample cell includes the correlation and autocorrelation between the state sequence of the sample cell and the second latent state of the trend feature of the previous sample cell in the current iteration.

[0106] That is, the embodiment of the present application will obtain the correlation and autocorrelation of adjacent battery cells based on seasonal characteristics and trend characteristics, thereby further improving the information mining of the state sequence of the sample battery cells.

[0107] Based on the above embodiments, as an optional embodiment, obtaining a first correlation between the seasonal characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration according to the seasonal characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration includes:

[0108] S501. For each sample battery cell, weight the seasonal characteristics of the state sequence of the sample battery cell in the current iteration by the first weight of the current iteration to obtain the first seasonal characteristics of the sample battery cell in the current iteration; weight the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration by the second weight of the current iteration to obtain the third hidden state of the previous sample battery cell in the current iteration.

[0109] S502a. For each sample battery cell, perform an autocorrelation operation based on the first seasonal feature of the sample battery cell in the current iteration, the third hidden state of the previous sample battery cell in the current iteration, and the first bias term in the current iteration to obtain an autocorrelation between the state sequence of the sample battery cell and the second hidden state of the seasonal feature of the previous sample battery cell in the current iteration.

[0110] Specifically, the embodiment of the present application can use Formula 4 to calculate the autocorrelation between the state sequence of the sample cell and the second hidden state of the seasonal characteristics of the previous sample cell in this round of iteration: :

[0111] Formula 4

[0112] in, represents the autocorrelation operation, Indicates sample battery cells t In this round of seasonal characteristics, Indicates sample battery cells t -1 is the second hidden state of the seasonal feature in this iteration, represents the first weight, represents the second weight, represents the first bias term.

[0113] S502b performs a correlation operation on each sample cell based on the first seasonal feature of the sample cell in the current iteration, the third hidden state of the previous sample cell in the current iteration, and the second bias term of the current iteration to obtain a correlation between the state sequence of the sample cell and the second hidden state of the seasonal feature of the previous sample cell in the current iteration.

[0114] Specifically, the embodiment of the present application can use Formula 5 to calculate the correlation between the state sequence of the sample battery and the second hidden state of the seasonal characteristics of the previous sample battery in this round of iteration: :

[0115] Formula 5

[0116] in, Represents a correlation operation, represents the second bias term.

[0117] Based on the above embodiments, as an optional embodiment, obtaining a second correlation between the trend feature of the state sequence of the sample battery cell in the current iteration and the second hidden state of the trend feature of the state sequence of the previous sample battery cell in the current iteration includes:

[0118] S601: For each sample cell, weight the trend feature of the state sequence of the sample cell in the current iteration by using the third weight of the current iteration to obtain the first trend feature of the sample cell in the current iteration; and weight the second hidden state of the trend feature of the state sequence of the previous sample cell in the current iteration by using the fourth weight of the current iteration to obtain the fourth hidden state of the previous sample cell in the current iteration.

[0119] S602a. For each sample cell, perform an autocorrelation operation based on the first trend feature of the sample cell in the current iteration, the fourth hidden state of the previous sample cell in the current iteration, and the third bias term in the current iteration to obtain an autocorrelation between the state sequence of the sample cell and the second hidden state of the trend feature of the previous sample cell in the current iteration.

[0120] Specifically, the embodiment of the present application can use Formula 6 to calculate the autocorrelation between the state sequence of the sample cell and the second hidden state of the trend feature of the previous sample cell in this round of iteration: :

[0121] Formula 6

[0122] in, Indicates sample battery cellst In this round of iteration trend characteristics, Indicates sample battery cells t -1 is the second hidden state of the trend feature in this round of iteration, represents the third weight, represents the fourth weight, represents the third bias term.

[0123] S602b. For each sample cell, perform a correlation operation based on the first trend feature of the sample cell in the current iteration, the fourth hidden state of the previous sample cell in the current iteration, and the fourth bias item in the current iteration to obtain a correlation between the state sequence of the sample cell and the second hidden state of the trend feature of the previous sample cell in the current iteration.

[0124] Specifically, the embodiment of the present application can use Formula 7 to calculate the correlation between the state sequence of the sample cell and the second hidden state of the trend feature of the previous sample cell in this round of iteration: :

[0125] Formula 7

[0126] in, Represents a correlation operation, represents the fourth bias term.

[0127] It should be noted that the second hidden state of the CGRU model in the embodiment of the present application is composed of the three-dimensional hidden state 、 、 Composition, of which and is a recursive update, It is not updated recursively, but by Updated by the input of the moment; The hidden state Use only The remaining part of and The update of is shown in formulas 8 and 9:

[0128] Formula 8

[0129] Formula 9

[0130] in, represents the fifth partial term, represents the sixth partial term, represents the ratio of the weighted sum (between 0 and 1), Represents dot product.

[0131] It should be noted that if There is a value close to 0, then will also have a value close to 0; therefore, if and The autocorrelation value between is low, then It will be assigned a high weight to update ; The same process applies if and The correlation between If the value of is close to 0, then The value of will also be close to 0; similarly, if and The correlation between It will be assigned a high weight to update .at last, Finally, update the weighted sum of the left and right terms of the above formula.

[0132] is a parameter that determines the ratio of the weighted sum of the two terms and has a value between 0 and 1; if If the value of is 0.5, then the same weighted sum is used to update these two terms.

[0133] Based on the above embodiments, as an optional embodiment, the predicted detection result of the sample battery cell can be obtained by formula 10:

[0134] Formula 10

[0135] The embodiment of the present application provides a device for detecting abnormal battery cells, such as Figure 5 As shown, the device may include: a test information acquisition module 501, a state sequence prediction module 502 and a detection model output module 503, wherein:

[0136] A test information acquisition module 501 is used to acquire test information of multiple cells to be tested, where each test information is a parameter set of the corresponding cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process;

[0137] A state sequence prediction module 502 is configured to input test data of a plurality of cells to be tested into a pre-trained state prediction model, and obtain a state sequence of the plurality of cells to be tested output by the state prediction model, wherein the state sequence includes a first latent state of the cells to be tested in each process, and each first latent state is used to indicate whether the corresponding cell to be tested is abnormal in the corresponding process;

[0138] a detection model output module 503, configured to input the state sequences of the plurality of cells to be tested into a pre-trained anomaly detection model according to the production order of the cells to be tested, and obtain detection results of the plurality of cells to be tested output by the anomaly detection model, wherein the detection results are used to indicate whether each cell to be tested is abnormal;

[0139] The state prediction model is formed by training a Hidden Markov Model (HMM) model using the test information of the sample battery cell as a training sample and the state sequence of the sample battery cell as a training label;

[0140] The anomaly detection model is formed by training a correlation gated recurrent (CGRU) model using a state sequence of multiple sample cells arranged in the production order of the sample cells as training samples and a detection result of multiple sample cells arranged in the production order of the sample cells as training labels.

[0141] The device of the embodiment of the present application can execute the method provided by the embodiment of the present application, and its implementation principle is similar. The actions performed by each module in the device of each embodiment of the present application correspond to the steps in the method of each embodiment of the present application. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, and will not be repeated here.

[0142] In an embodiment of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory. The processor executes the above-mentioned computer program to implement the steps of a method for detecting abnormal battery cells. Compared with the related art, it can be achieved: first, test information of multiple battery cells to be tested is obtained, each test information is a parameter set of the corresponding battery cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process. The test data of the multiple battery cells to be tested are input into a pre-trained state prediction model to obtain the state sequence of the multiple battery cells to be tested output by the state prediction model. The state sequence of each of the multiple battery cells to be tested is continued to be used as the input of the abnormality detection model. The abnormality detection model learns the autocorrelation within the time series constituted by the state sequence of the multiple battery cells to be tested and the correlation between different features, and finally obtains the detection result to perform abnormal screening on the battery cells, thereby improving the detection efficiency and detection accuracy.

[0143] In an alternative embodiment, an electronic device is provided, such as Figure 6 As shown, Figure 6The electronic device 4000 shown includes: a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which may be used for data exchange between the electronic device and other electronic devices, such as data transmission and / or data reception. It should be noted that in actual applications, the number of transceivers 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present application.

[0144] Processor 4001 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 4001 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0145] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, for example. Bus 4002 may be divided into an address bus, a data bus, a control bus, and so on. For ease of illustration, bus 4002 is represented by a single thick line in the figure, but this does not indicate that there is only one bus or only one type of bus.

[0146] The memory 4003 may be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, without limitation herein.

[0147] The memory 4003 is used to store the computer program for executing the embodiment of the present application, and the execution is controlled by the processor 4001. The processor 4001 is used to execute the computer program stored in the memory 4003 to implement the steps shown in the above method embodiment.

[0148] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0149] An embodiment of the present application also provides a computer program product, including a computer program, which can implement the steps and corresponding contents of the aforementioned method embodiment when executed by a processor.

[0150] In the specification and claims of this application and the accompanying drawings, the terms "first," "second," "third," "fourth," "1," "2," and so on (if any) are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that the terms used in this manner are interchangeable where appropriate, such that the embodiments of the present application described herein can be practiced in an order other than that shown or described.

[0151] It should be understood that, although each operation step is indicated by arrows in the flowchart of the embodiment of the present application, the order of implementation of these steps is not limited to the order indicated by the arrows. Unless otherwise clearly stated herein, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be performed in other orders according to demand. In addition, some or all of the steps in each flowchart can include multiple sub-steps or multiple stages based on actual implementation scenarios. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage in these sub-steps or stages can also be executed at different times respectively. Under different scenarios at the execution time, the execution order of these sub-steps or stages can be flexibly configured according to demand, and the embodiment of the present application does not limit this.

[0152] The above description is only an optional implementation method for some implementation scenarios of this application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of this application, the use of other similar implementation methods based on the technical ideas of this application also falls within the protection scope of the embodiments of this application.

Claims

1. A method for detecting abnormal battery cells, characterized in that: include: Acquire test information of multiple battery cells to be tested, where each piece of test information is a parameter set of the corresponding battery cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process; Inputting test data of multiple cells to be tested into a pre-trained state prediction model to obtain a state sequence of multiple cells to be tested output by the state prediction model, wherein the state sequence includes a first latent state of the cell to be tested in each process, and each first latent state is used to indicate whether the corresponding cell to be tested is abnormal in the corresponding process; Inputting the state sequences of the plurality of cells to be tested into a pre-trained anomaly detection model according to the production order of the cells to be tested, and obtaining detection results of the plurality of cells to be tested output by the anomaly detection model, wherein the detection results are used to indicate whether each cell to be tested is abnormal; The state prediction model is formed by training a Hidden Markov Model (HMM) using the test information of the sample battery cell as a training sample and the state sequence of the sample battery cell as a training label; The anomaly detection model is formed by training a correlation gated recurrent (CGRU) model using a state sequence of multiple sample cells arranged in the production order of the sample cells as training samples and a detection result of multiple sample cells arranged in the production order of the sample cells as training labels.

2. The method according to claim 1, characterized in that The state prediction model is trained in the following way: Initialize the initial state probability matrix, state transition probability matrix and emission state probability matrix of the HMM model; Constructing an observation sequence, wherein the observation sequence is test information of the sample battery cell; The HMM model is trained and parameter optimized using the observation sequence, and the corresponding relationship between the observation sequence and the training label and the rule of mutual transition change of the first hidden states of adjacent processes are obtained, and the optimal state transition probability matrix and emission state probability matrix are output; An optimal HMM model is obtained based on the optimal state transition probability matrix and the emission state probability matrix as the state prediction model.

3. The method according to claim 1, characterized in that The anomaly detection model is trained in the following way: The state sequences of multiple sample cells arranged in the order of their production are input into the CGRU model for time series analysis and correlation analysis to obtain the residual features of the state sequence of each sample cell, the second latent state of the seasonal features, and the second latent state of the trend features. According to the production order of the sample cells and the residual features of the state sequence of each sample cell, the second latent state of the seasonal feature, and the second latent state of the trend feature, the prediction detection results of each sample cell are obtained in sequence; According to the predicted test results of each sample battery cell and the difference between the test results, the CGRU model is trained to meet the preset conditions to obtain the anomaly detection model.

4. The method according to claim 3, characterized in that The state sequences of the plurality of sample cells arranged in the production order of the sample cells are input into the CGRU model for time series analysis and correlation analysis to obtain the residual features, the second latent state of the seasonal features, and the second latent state of the trend features of each sample cell, including: The state sequences of multiple sample cells arranged in the production order of the sample cells are input into the CGRU model for multiple rounds of iterations; Among them, for each round of iteration: Perform time series analysis on the state sequence of each sample battery cell according to the production order of the sample battery cells to obtain the residual characteristics, seasonal characteristics, and trend characteristics of the state sequence of each sample battery cell in this round of iteration; For each sample battery cell, based on the seasonal characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration, obtain a first correlation degree between the seasonal characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration, as the first correlation degree corresponding to the sample battery cell; For each sample battery cell, based on the trend feature of the state sequence of the sample battery cell in the current iteration and the second hidden state of the trend feature of the state sequence of the previous sample battery cell in the current iteration, obtain a second correlation degree between the trend feature of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal feature of the state sequence of the previous sample battery cell in the current iteration, as the second correlation degree corresponding to the sample battery cell; For each sample cell, obtaining the second hidden state of the seasonal characteristics of the sample cell in the current iteration according to the first correlation degree corresponding to the sample cell, the seasonal characteristics in the current iteration, and the second hidden state of the seasonal characteristics of the previous sample cell in the current iteration; For each sample cell, the second hidden state of the trend feature of the sample cell in the current iteration is obtained according to the second correlation degree corresponding to the sample cell, the trend feature in the current iteration, and the second hidden state of the trend feature of the previous sample cell in the current iteration.

5. The method according to claim 4, characterized in that The first correlation degree of each sample cell includes the correlation and autocorrelation between the state sequence of the sample cell and the second hidden state of the seasonal characteristics of the previous sample cell in this iteration; The second correlation degree of each sample cell includes the correlation and autocorrelation between the state sequence of the sample cell and the second latent state of the trend feature of the previous sample cell in the current iteration.

6. The method according to claim 5, characterized in that The obtaining, based on the seasonal characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration, a first correlation degree between the seasonal characteristics of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal characteristics of the state sequence of the previous sample battery cell in the current iteration includes: For each sample cell, the seasonal characteristics of the state sequence of the sample cell in the current iteration are weighted by the first weight of the current iteration to obtain the first seasonal characteristics of the sample cell in the current iteration. The second hidden state of the seasonal characteristics of the state sequence of the previous sample cell in the current iteration is weighted by the second weight of the current iteration to obtain the third hidden state of the previous sample cell in the current iteration. For each sample cell, an autocorrelation operation is performed based on the first seasonal feature of the sample cell in the current iteration, the third hidden state of the previous sample cell in the current iteration, and the first bias term of the current iteration to obtain the autocorrelation between the state sequence of the sample cell and the second hidden state of the seasonal feature of the previous sample cell in the current iteration; For each sample cell, a correlation operation is performed based on the first seasonal feature of the sample cell in this iteration, the third hidden state of the previous sample cell in this iteration, and the second bias term in this iteration to obtain the correlation between the state sequence of the sample cell and the second hidden state of the seasonal feature of the previous sample cell in this iteration.

7. The method according to claim 5, characterized in that Obtaining, based on the trend feature of the state sequence of the sample battery cell in the current iteration and the second hidden state of the trend feature of the state sequence of the previous sample battery cell in the current iteration, a second correlation degree between the trend feature of the state sequence of the sample battery cell in the current iteration and the second hidden state of the seasonal feature of the state sequence of the previous sample battery cell in the current iteration includes: For each sample cell, weight the trend feature of the state sequence of the sample cell in the current iteration by the third weight of the current iteration to obtain the first trend feature of the sample cell in the current iteration; weight the second hidden state of the trend feature of the state sequence of the previous sample cell in the current iteration by the fourth weight of the current iteration to obtain the fourth hidden state of the previous sample cell in the current iteration; For each sample cell, an autocorrelation operation is performed based on the first trend feature of the sample cell in the current iteration, the fourth hidden state of the previous sample cell in the current iteration, and the third bias term in the current iteration to obtain the autocorrelation between the state sequence of the sample cell and the second hidden state of the trend feature of the previous sample cell in the current iteration; For each sample cell, a correlation operation is performed based on the first trend feature of the sample cell in the current iteration, the fourth hidden state of the previous sample cell in the current iteration, and the fourth bias term in the current iteration to obtain the correlation between the state sequence of the sample cell and the second hidden state of the trend feature of the previous sample cell in the current iteration.

8. The method according to any one of claims 1 to 7, characterized in that The test parameter in the parameter set is at least one initial test parameter with the highest evaluation score screened out after evaluating multiple initial test parameters using a preset scoring function.

9. A device for detecting abnormal battery cells, characterized in that: include: A test information acquisition module is used to obtain test information of multiple cells to be tested, each of which is a parameter set of the corresponding cell to be tested under multiple processes, and each parameter set includes at least one test parameter of the corresponding process; A state sequence prediction module is used to input the test data of multiple cells to be tested into a pre-trained state prediction model to obtain a state sequence of the multiple cells to be tested output by the state prediction model, wherein the state sequence includes the first latent state of the cells to be tested in each process, and each first latent state is used to indicate whether the corresponding cell to be tested is abnormal in the corresponding process; a detection model output module, configured to input the state sequences of the plurality of cells to be tested into a pre-trained anomaly detection model according to the production order of the cells to be tested, and obtain detection results of the plurality of cells to be tested output by the anomaly detection model, wherein the detection results are used to indicate whether each cell to be tested is abnormal; The state prediction model is formed by training a Hidden Markov Model (HMM) model using the test information of the sample battery cell as a training sample and the state sequence of the sample battery cell as a training label; The anomaly detection model is formed by training a correlation gated recurrent (CGRU) model using a state sequence of multiple sample cells arranged in the production order of the sample cells as training samples and a detection result of multiple sample cells arranged in the production order of the sample cells as training labels.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the abnormal battery cell detection method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the abnormal battery cell detection method according to any one of claims 1 to 8 is implemented.

12. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the abnormal battery cell detection method according to any one of claims 1 to 8 is implemented.

Images