CN-122017596-A - Method and system for detecting abnormity of K value of battery cell

CN122017596ACN 122017596 ACN122017596 ACN 122017596ACN-122017596-A

Abstract

The invention discloses a method and a system for detecting abnormal K value of a battery core, and relates to the technical field of lithium ion battery manufacturing and detection; the method comprises the steps of reconstructing open-circuit voltage data into a time sequence, splicing the time sequence with production process data to form a multidimensional input feature vector, inputting the feature vector into a pre-trained GRU-BiLSTM mixed prediction model, and outputting a prediction K value and a classification judgment result. The invention realizes accurate prediction and early abnormality identification of the K value of the battery cell, shortens the detection period to 3-5 days, improves the abnormality classification accuracy to more than 95%, can rapidly locate the abnormality source, and obviously improves the efficiency, quality control and traceability of lithium battery production.

Inventors

CHEN GUANGHUA
WANG YAN

Assignees

合肥国轩高科动力能源有限公司

Dates

Publication Date: 20260512
Application Date: 20260228

Claims (10)

1. The method for detecting the abnormal K value of the battery cell is characterized by comprising the following steps of: S1, collecting production process data of a target battery cell and open circuit voltage data at a plurality of set time points within preset standing time, wherein the production process data at least comprises one of pole piece compaction density, liquid injection amount, formation voltage, formation time, pole piece baking temperature and pole lug welding strength; S2, reconstructing the open-circuit voltage data into an open-circuit voltage time sequence, and splicing the open-circuit voltage time sequence with the production process data to form a multidimensional input feature vector; S3, inputting the multidimensional input feature vector into a pre-trained mixed prediction model, wherein the mixed prediction model is used for extracting the time sequence feature of the open-circuit voltage time sequence, fusing the time sequence feature with the production process data, and outputting a predicted K value of the target battery cell and a classification judgment result about whether the K value is abnormal or not based on the fused feature.
2. The method for detecting abnormal K value of battery cell according to claim 1, wherein the hybrid prediction model in step S3 includes an input layer, a GRU layer, a BiLSTM layer, a feature fusion layer and an output layer connected in sequence, wherein, The input layer is used for receiving the multidimensional input feature vector, inputting the open-circuit voltage time sequence into the GRU layer, and transmitting the production process data to the feature fusion layer; the GRU layer adopts a ReLU activation function and is used for extracting short-term time sequence characteristics of the open-circuit voltage time sequence; The BiLSTM layers adopt a tanh activation function for extracting long-term timing characteristics of the open-circuit voltage timing sequence from forward and reverse directions based on the short-term timing characteristics; the feature fusion layer comprises two layers of fully-connected networks, and a ReLU activation function is adopted for splicing and fusing the long-term time sequence features and the production process data; The output layer comprises a first output unit and a second output unit, wherein the first output unit adopts a Linear activation function and is used for generating the predicted K value based on the fused features output by the feature fusion layer, and the second output unit adopts a Sigmoid activation function and is used for obtaining the classification judgment result through a preset classification probability threshold value based on the predicted K value.
3. The method for detecting abnormal cell K value according to claim 2, wherein the joint loss function based on a combination of MSE loss and cross entropy loss Training the hybrid predictive model by training data, wherein the training data is a multi-dimensional input feature vector set with labels, the labels comprise real K value labels and real classification labels, and the joint loss function The formula of (2) is as follows: Wherein, the Losing weight for MSE; As a function of the loss of the MSE, Wherein N is the total number of the battery cells to be detected, As a true K-value tag, Is the predicted K value; In order to cross-entropy loss function, Wherein N is the total number of the battery cells to be detected, For a true class of labels, To predict classification probability.
4. The method for detecting abnormality of a cell K value according to claim 1, wherein the classification determination result is made based on at least one of: the predicted K value output by the output layer of the hybrid prediction model exceeds a preset dynamic threshold; and the classification probability output by the output layer of the mixed prediction model is larger than the preset classification probability threshold.
5. The method for detecting abnormal cell K value according to claim 4, wherein the dynamic threshold is set by: dividing the electric core into different types of groups according to the characteristics affecting the K value attenuation rule of the electric core, wherein the characteristics comprise an electric core chemical system of the electric core; For each cell type group, collecting open circuit voltage data and corresponding K value data of a plurality of set time points of the historical normal cells in a preset standing time; Based on the K value data corresponding to each cell type group, analyzing to obtain K value attenuation thresholds at different set time points; and establishing a mapping relation between preset standing time and a K value attenuation threshold value, and generating a dynamic threshold value of the type group.
6. The method for detecting abnormal K value of a battery cell according to claim 1, further comprising: when the classification judgment result is abnormal, extracting a multidimensional input feature vector corresponding to the abnormal battery cell; Calculating SHAP values of the dimensional data in the multidimensional input feature vector by using a SHAP interpretability algorithm; Determining data with top ranking on the K value abnormal influence weight based on the absolute value ranking of the SHAP values; And generating a traceability report indicating the source of the abnormality by combining production process standards corresponding to the data.
7. The method according to any one of claims 1 to 6, wherein in the step S1, the preset rest time is 1 hour to 168 hours after the start of rest, and the set time points are not less than three.
8. The method according to claim 7, wherein the collected production process data and open circuit voltage data are cleaned before the step S2, and the cleaned production process data are subjected to standardized processing.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable by the processor, characterized in that the processor implements the cell K value anomaly detection method according to any one of claims 1 to 8 when the computer program is executed.
10. The system for detecting the abnormal K value of the battery cell is characterized by comprising a cloud layer, an edge layer and a terminal layer, wherein, The terminal layer is deployed in a battery cell production workshop and is configured to collect original data, wherein the original data comprises production process data of a target battery cell and open-circuit voltage data at a plurality of set time points in preset standing time; the edge layer is deployed on a workshop local server and comprises the electronic equipment described in claim 9, wherein the electronic equipment is used for executing a cell K value abnormality detection method based on the original data of the terminal layer to generate a predicted K value and an abnormality judgment result, and outputting an early warning signal and a tracing report when the cell K value is abnormal; the cloud layer is deployed on a cloud platform and is configured to retrain and update the hybrid prediction model based on historical data received from the electronic device and send the updated hybrid prediction model to the electronic device.

Description

Method and system for detecting abnormity of K value of battery cell Technical Field The invention relates to the technical field of lithium ion battery manufacturing and detection, in particular to a method and a system for detecting abnormal K value of a battery cell. Background In the lithium battery manufacturing field, the K value (self-discharge rate) of the battery core is a core index for measuring the storage performance and the cycle life of the battery core, and abnormal K value (such as sudden rise of the K value) can directly lead to rapid attenuation and poor consistency of the capacity of the battery core, and even lead to safety accidents. The current detection mode of the K value of the main stream in the industry is standing-Open Circuit Voltage (OCV) detection, namely, firstly placing the battery cell in a constant temperature environment of 25 ℃ for standing for 7-14 days, then calculating the K value (K=delta OCV/delta t) by the difference value of the OCVs before and after standing, and comparing with a fixed threshold value (less than or equal to 0.05 mV/h) to judge whether the battery cell is abnormal. In order to improve the detection efficiency, some enterprises try to shorten the standing time or increase the detection frequency, but the method has technical limitations that firstly, the traditional method relies on manual reading of OCV data and calculation of K value, the detection efficiency is low, misjudgment is easily caused by manual misoperation, secondly, abnormality is judged only by means of 'single time point OCV difference', dynamic attenuation trend of OCV in the standing process is not considered, and cell production process data is not associated, so that the abnormality judgment accuracy is only 75% -85%, a large number of potential abnormal cells flow into a downstream ring section, after-sales risk is improved, thirdly, pole piece compaction density, liquid injection amount, formation temperature and other data generated in the production process and standing multi-time point OCV data are only archived, an associated analysis model with the K value is not established, the data utilization rate is low, support cannot be provided for abnormality sources, fourthly, the detection period is long, the state of the K value can be judged after the whole period standing is required, the standing period of 7-14 days cannot adapt to large-scale production requirements of lithium battery, and the serious influence on the delivery period of the lithium battery industry is caused if the K value occurs. Disclosure of Invention Aiming at the problems of lower detection efficiency, lower abnormality judgment accuracy and lower data utilization rate and longer detection period of the detection of the K value of the battery cell in the prior art, the invention provides the detection method and the detection system of the K value of the battery cell. According to one aspect of the invention, the method for detecting the abnormal K value of the battery cell comprises the following steps of S1, collecting production process data of a target battery cell and open-circuit voltage data at a plurality of set time points within preset standing time, wherein the production process data at least comprise one of pole piece compaction density, liquid injection amount, formation voltage, formation time, pole piece baking temperature and pole lug welding strength, S2, reconstructing the open-circuit voltage data into an open-circuit voltage time sequence, splicing the open-circuit voltage time sequence with the production process data to form a multidimensional input feature vector, S3, inputting the multidimensional input feature vector into a pre-trained mixed prediction model, wherein the mixed prediction model is used for extracting time sequence features of the open-circuit voltage time sequence, fusing the time sequence features with the production process data, and outputting a predicted K value of the target battery cell and a classification determination result about whether the K value is abnormal or not based on the fused features. According to the above aspect, more specifically, the hybrid prediction model in the step S3 includes an input layer, a GRU layer, a BiLSTM layer, a feature fusion layer and an output layer which are sequentially connected, wherein the input layer is used for receiving the multidimensional input feature vector, inputting the open-circuit voltage time sequence therein into the GRU layer, and transmitting the production process data to the feature fusion layer, the GRU layer adopts a ReLU activation function for extracting the short-term time sequence feature of the open-circuit voltage time sequence, the BiLSTM layer adopts a tanh activation function for extracting the long-term time sequence feature of the open-circuit voltage time sequence from the forward direction and the reverse direction based on the short-term time sequence feature, the feature fusio