Battery short circuit detection system and device fusing impedance spectroscopy and interpretable ensemble learning

Abstract

The invention belongs to the field of batteries, and provides a battery short circuit detection system and device fusing impedance spectroscopy and interpretable ensemble learning. The system comprises a multi-dimensional impedance feature extraction module; a first layer model training module; a second layer element learning device training module; the key frequency band feature screening module is used for carrying out importance analysis of local features and global features on the trained double-layer integrated model based on an SHAP interpretability analysis framework, identifying contribution degrees of impedance features in internal short circuit detection under different frequencies, and screening out key frequency band features; and the model retraining module is used for retraining the first-layer integrated model and the second-layer element learning device based on the important frequency band characteristics, and carrying out battery short circuit detection through the retrained and optimized first-layer integrated model and second-layer element learning device. High-precision detection and mechanism analysis of the short-circuit fault in the battery are realized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of batteries, and particularly relates to a battery short circuit detection system and device fusing impedance spectrum and explainable ensemble learning. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] During long-term use, the battery is prone to internal short circuit failure due to factors such as electrolyte decomposition, separator damage, and active material precipitation, which seriously threatens the safety and reliability of the battery system. The traditional battery internal short circuit detection method mainly relies on the monitoring of voltage, current, temperature and other macroscopic parameters, but the method based on voltage and current monitoring can only detect obvious internal short circuit failure, and is not sensitive to early and slight internal short circuit phenomenon, which is prone to missed detection; the temperature monitoring based method has a lagging response, and when the temperature anomaly is detected, the internal short circuit failure has often developed to a more serious stage; the detection method based on single sensor data has poor anti-interference ability and is prone to false positives under complex working conditions.

[0004] In order to solve the above problems, the electrochemical impedance spectroscopy (EIS) technology can reflect the changes of the electrochemical process and interface characteristics inside the battery, and provides a new technical approach for early detection of internal short circuit failure. However, the existing EIS-based detection methods still have the following shortcomings: 1) The traditional method mainly uses a single machine learning algorithm, and the model has limited generalization ability, and the detection accuracy is unstable under different types of batteries or different working conditions; the battery failure data is difficult to collect, and in actual application, it often faces the challenge of small sample training data set, and the performance of traditional machine learning method significantly decreases under small sample condition, which is difficult to meet the actual engineering requirements.

[0005] 2) The existing method lacks explainable analysis of the detection results, and cannot provide in-depth insight into the failure mechanism, which limits the practicability of the method; and the feature extraction method is relatively simple, and cannot fully extract the potential feature information in the multi-frequency point impedance data. SUMMARY

[0006] In order to solve the above technical problems, the present application provides a battery short circuit detection system and device fusing impedance spectrum and explainable ensemble learning, which realizes high-precision detection and mechanism analysis of battery internal short circuit failure through double-layer ensemble learning architecture and SHAP explainable analysis.

[0007] In order to achieve the above purpose, the present application adopts the following technical solutions: The first aspect of the present application provides a battery short circuit detection system fusing impedance spectrum and explainable ensemble learning.

[0008] In one or more embodiments, a battery short circuit detection system integrating impedance spectrum and explainable ensemble learning comprises: a multi-dimensional impedance feature extraction module for obtaining electrochemical impedance spectrum data of a battery at multiple frequency points, extracting impedance real part and imaginary part features at different frequencies, and then obtaining a standardized multi-dimensional impedance feature dataset after standardization preprocessing; a first layer model training module for training a first layer ensemble model based on the standardized multi-dimensional impedance feature dataset, and obtaining a prediction dataset output by the first layer model; a second layer meta-learner training module for performing feature fusion on the prediction probability dataset output by the first layer model and the standardized multi-dimensional impedance feature dataset to form an enhanced feature dataset, and training a second layer meta-learner based on the enhanced feature dataset, thereby forming a double-layer ensemble model composed of the first layer ensemble model and the second layer meta-learner; a key frequency band feature screening module for performing importance analysis on local features and global features of the trained double-layer ensemble model based on a SHAP explainability analysis framework, identifying the contribution of each impedance feature at different frequencies to internal short circuit detection, and screening key frequency band features; a model retraining module for retraining the first layer ensemble model and the second layer meta-learner based on the important frequency band features, and performing battery short circuit detection by using the retrained and optimized first layer ensemble model and second layer meta-learner.

[0009] As an implementation, the process of performing importance analysis on local features and global features of the trained double-layer ensemble model based on the SHAP explainability analysis framework is as follows: a SHAP interpreter is constructed based on the trained double-layer ensemble model, and the impedance real part and imaginary part features extracted at the original multiple frequency points are taken as the explanation objects; the contribution of the impedance real part and imaginary part features is quantified using the Shapley value theory, and the SHAP values of each impedance feature are obtained; for any test sample, the SHAP value set of all impedance features of the sample is calculated to obtain the local feature importance distribution of the sample; the average absolute value of the SHAP values of each feature is calculated on all training samples to obtain the global importance index of a certain impedance feature.

[0010] As an implementation, in the process of screening key frequency band features, the global importance index of a certain impedance feature is used to globally sort the impedance features at all frequencies, and the impedance features of the top k most important frequency points or corresponding frequency intervals are selected as the key frequency band features; wherein k is a positive integer greater than or equal to 2.

[0011] As an implementation form, the first layer integrated model is trained based on the standardized multi-dimensional impedance feature dataset by adopting a K-fold cross-validation strategy.

[0012] As an implementation form, the first layer integrated model includes four heterogeneous base learners of XGBoost, CatBoost, random forest and TabPFN, and the prediction probability output on the training set is obtained through cross-validation.

[0013] As an implementation form, the prediction probability dataset output by the first layer model and the standardized multi-dimensional impedance feature dataset are fused by adopting a horizontal splicing strategy to obtain an enhanced feature dataset.

[0014] As an implementation form, in the process of battery short circuit detection by retraining the optimized first layer integrated model and the second layer meta-learner, the probability output by the second layer meta-learner is compared with a preset threshold to make a binary classification decision, and whether there is a short circuit fault in the battery is detected.

[0015] The second aspect of the application provides a battery short circuit detection method fusing impedance spectrum and interpretable integrated learning.

[0016] In one or more embodiments, a battery short circuit detection method fusing impedance spectrum and interpretable integrated learning is provided, comprising: Obtaining electrochemical impedance spectrum data of a battery at multiple frequency points, extracting impedance real part and imaginary part features at different frequencies, and then obtaining a standardized multi-dimensional impedance feature dataset through standardization preprocessing; Training a first layer integrated model based on the standardized multi-dimensional impedance feature dataset to obtain a prediction dataset output by the first layer model; Performing feature fusion on the prediction probability dataset output by the first layer model and the standardized multi-dimensional impedance feature dataset to form an enhanced feature dataset, and training a second layer meta-learner based on the enhanced feature dataset to form a double-layer integrated model composed of the first layer integrated model and the second layer meta-learner; Based on the SHAP interpretability analysis framework, the importance of local features and global features of the trained double-layer integrated model is analyzed to identify the contribution of each impedance feature at different frequencies to the internal short circuit detection, and key frequency band features are screened out; Based on the important frequency band features, the first layer integrated model and the second layer meta-learner are retrained, and the battery short circuit detection is performed by the retrained and optimized first layer integrated model and the second layer meta-learner.

[0017] The third aspect of the application provides a computer readable storage medium.

[0018] A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the battery short circuit detection method fusing impedance spectrum and explainable integrated learning as described above.

[0019] A fourth aspect of the present application provides an electronic device.

[0020] An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the battery short circuit detection method fusing impedance spectrum and explainable integrated learning as described above when executing the program.

[0021] Compared with the prior art, the present application has the following beneficial effects: (1) The present application uses a first layer integrated model and a second layer meta-learner to construct a double-layer integrated model, the first layer integrated model is trained with a standardized multi-dimensional impedance feature data set, and the second layer meta-learner is trained with a feature data set in which the standardized multi-dimensional impedance feature data set and the output probability of the first layer integrated model are fused. Since the second layer meta-learner further fuses the output probability feature information of the first layer integrated model, the generalization ability and robustness of the detection model are significantly improved, and the accuracy of the detection result is improved.

[0022] (2) The present application forms a standardized multi-dimensional impedance feature through comprehensive collection and feature extraction of multi-frequency point electrochemical impedance spectrum data, so that early feature changes of the battery internal short circuit fault can be captured, and timely detection of the fault can be realized.

[0023] (3) The present application introduces a SHAP explainability analysis framework to analyze the importance of local features and global features of the trained double-layer integrated model, identify the contribution of each impedance feature at different frequencies to the internal short circuit detection, screen key frequency band features, and retrain the double-layer integrated model using the key frequency band features. This not only provides accurate detection results, but also quantifies the contribution of impedance features at different frequencies to the detection results, providing a scientific basis for fault mechanism analysis and model optimization. BRIEF DESCRIPTION OF DRAWINGS

[0024] The drawings accompanying the specification of the present application form a part thereof, serve to provide further understanding of the present application, and together with the description, explain the present application. The present application is shown by way of illustration in the drawings and their description is not intended as an improper limitation on the present application.

[0025] Figure 1 is a flowchart of the battery short circuit detection method fusing impedance spectrum and explainable integrated learning of the embodiment of the present application; Figure 2 is a double-layer integrated model of the embodiment of the present application; Figure 3This is a schematic diagram of the battery short-circuit detection system based on the fusion of impedance spectrum and interpretable integrated learning, according to an embodiment of the present invention. Figure 4 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0028] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0029] Figure 1 This is a flowchart illustrating a battery short-circuit detection method that integrates impedance spectroscopy and interpretable ensemble learning, as described in an embodiment of the present invention. Figure 1 The battery short-circuit detection method based on fused impedance spectrum and interpretable integrated learning in this embodiment may include: S101: Obtain electrochemical impedance spectroscopy (EIS) data of the battery at multiple frequency points, extract the real and imaginary impedance features at different frequencies, and then perform standardized preprocessing to obtain a standardized multidimensional impedance feature dataset.

[0030] Specifically, the process of S101 is as follows: (1) Apply a small-amplitude sinusoidal perturbation signal while the battery is in normal working condition; (2) Collect current and voltage response data at both ends of the battery within a preset frequency range; (3) Perform frequency domain transformation on the acquired current and voltage data to calculate the complex impedance data: ; (4) Extract the real part of the impedance based on the obtained complex impedance data. and the imaginary part of impedance Construct the original multidimensional impedance eigenvector:

[0031] in, For the first a test frequency point, is the total number of frequency points.

[0032] (5) The extracted impedance characteristics are standardized for pretreatment, and the standardization formula is:

[0033] wherein, is the original impedance characteristic vector, is the mean vector of the training set characteristics, is the standard deviation vector of the training set characteristics, is the standardized characteristic vector.

[0034] (6) Constructing a standardized multi-dimensional impedance characteristic data set .

[0035] S102, based on the standardized multi-dimensional impedance characteristic data set, training the first layer integrated model to obtain the prediction data set output by the first layer model.

[0036] In the embodiments of the present application, the first layer integrated model includes four kinds of heterogeneous base learners of XGBoost, CatBoost, random forest and TabPFN, and the prediction probability output on the training set is obtained through cross-validation.

[0037] It should be noted here that in other embodiments, the first layer integrated model can also use other numbers of integrated heterogeneous base learners, and those skilled in the art can set it according to the actual situation, which will not be described in detail here.

[0038] In the embodiments of the present application, based on the standardized multi-dimensional impedance characteristic data set, the first layer integrated model is trained by using K-fold cross-validation strategy. In this way, through K-fold cross-validation and feature fusion strategy, the problem of overfitting is effectively avoided, and the stability and reliability of the model in actual application are improved.

[0039] Based on the standardized multi-dimensional impedance characteristic data set, the first layer integrated model is trained by using K-fold cross-validation strategy. The specific steps for obtaining the prediction probability output on the training set through cross-validation are as follows: S1021: The standardized multi-dimensional impedance characteristic data set is divided into K layers, and the value of K is usually set to 5-10 to ensure that the positive and negative sample ratios in each fold are consistent; S1022: For the kth fold cross-validation, the data set is divided into a training set and a validation set, and four kinds of heterogeneous base learners of XGBoost, CatBoost, random forest and TabPFN are trained on the training set; S1023: XGBoost, CatBoost and random forest models adopt Bayesian optimization for hyperparameter tuning, and the optimization objectives include but are not limited to learning rate, tree depth, regularization parameter, etc.; the TabPFN model is run based on the pre-training structure and adopts the default configuration; S1024: Using the trained four base learners to predict the validation set to obtain the prediction probability matrix ; S1025: Repeat steps S1022-S1024, complete K-fold cross-validation, rearrange all validation set prediction results in the original data order to obtain the complete training set prediction probability matrix .

[0040] S103, the prediction probability data set output by the first layer model is fused with the standardized multi-dimensional impedance feature data set to form an enhanced feature data set, and the second layer meta-learner is trained according to the enhanced feature data set, and then a double-layer integrated model composed of the first layer integrated model and the second layer meta-learner is formed, as shown in Figure 2 .

[0041] Specifically, the prediction probability data set output by the first layer model is fused with the standardized multi-dimensional impedance feature data set by using a horizontal splicing strategy to obtain an enhanced feature data set.

[0042] The prediction probability matrix output by the first layer model is fused with the standardized impedance feature matrix , and an enhanced feature matrix = [ ; The enhanced feature matrix is used as the input of the second layer TabPFN meta-learner to train the final detection model.

[0043] Herein, in the training of the double-layer integrated model, the optimal classification threshold is determined by but not limited to ROC curve analysis to balance the detection accuracy and false positive rate.

[0044] In the embodiments of the present application, TabPFN is used as the key learner of the second layer meta-learner, and the prior knowledge obtained by pre-training of TabPFN can effectively handle the battery fault detection under the condition of only a small amount of training samples, and still achieve high-precision internal short-circuit detection, effectively solving the actual problem of battery fault sample scarcity, and still achieving high-precision internal short-circuit detection under the condition of limited training data, significantly improving the practicality and engineering application value of the method.

[0045] In S104, based on the SHAP explainability analysis framework, importance analysis of local features and global features of the trained double-layer integrated model is performed to identify the contribution of each impedance feature to the internal short circuit detection at different frequencies and to screen out key frequency band features.

[0046] Specifically, based on the SHAP explainability analysis framework, the process of importance analysis of local features and global features of the trained double-layer integrated model is as follows: In S1041, a SHAP interpreter is constructed based on the trained double-layer integrated model, and the impedance real part and imaginary part features extracted at the original multi-frequency points are taken as the explanation objects. In this way, the intermediate prediction probability features introduced in the fusion stage are ignored, which can ensure that the explanation result is directly related to the actual impedance characteristics. In S1042, the contribution of the impedance real part and imaginary part features is quantified using the Shapley value theory to obtain the SHAP value of each impedance feature. The Shapley value theory is used to quantify the feature contribution. For any sample x, the SHAP value of the i-th feature is The calculation formula is as follows:

[0047] Where N represents the full set of all features, S is any subset that does not contain feature i, f(s) is the prediction output of the model on the feature subset S, represents the marginal contribution value of feature i in the current sample prediction.

[0048] In S1043, for any test sample, the SHAP value set of all impedance features is calculated to obtain the local feature importance distribution of the sample. For any j-th test sample , the SHAP value set of all impedance features is calculated to obtain the local feature importance distribution of the sample:

[0049] Where is the local feature importance distribution of sample , and represents the positive or negative influence of the i-th impedance feature on the classification output of sample , which supports traceability and physical explainability analysis of individual detection results. In S1044, the average absolute value of the SHAP value of each feature is calculated on all training samples to obtain the global importance index of a certain impedance feature.

[0050]

[0051] Where, is the feature importance value of the i-th feature, and N is the total number of training samples.

[0052] In the process of screening the key frequency band features, all impedance features at different frequencies are globally sorted according to the numerical value of the global importance indicator of a certain impedance feature, and the impedance features of the top k most important frequency points or corresponding frequency intervals are selected as the key frequency band features; wherein k is a positive integer greater than or equal to 2.

[0053] Specifically, all impedance features at different frequencies are globally sorted according to the numerical value of the global importance indicator of a certain impedance feature, and the impedance features of the top k most important frequency points or corresponding frequency intervals are selected:

[0054]

[0055] S105, based on the important frequency band features, retrain the first layer integrated model and the second layer meta-learner, and perform battery short circuit detection through the retrained and optimized first layer integrated model and the second layer meta-learner.

[0056] In the process of battery short circuit detection through the retrained and optimized first layer integrated model and the second layer meta-learner, the probability output by the second layer meta-learner is compared with a preset threshold to make a binary classification decision, and it is detected whether there is a short circuit fault in the battery.

[0057] The specific process of S105 is: S1051: based on the optimized feature set obtained in step S104 , extract the corresponding feature subset from the original training data set:

[0058] is the screened feature matrix; S1052: use the optimized training data set , retrain the double-layer integrated model according to steps S2-S3.

[0059] S1053: for the battery to be detected, collect EIS impedance data according to the process of step S101, but only extract the specific frequency point impedance features corresponding to the optimized feature set , and perform standardization preprocessing to obtain . S1054: input the standardized multi-dimensional impedance feature test data​​​​​​ Input the first layer of the trained four base learners, respectively, to obtain the prediction probability vector of each learner for the test data .

[0060] S1055: The test prediction probability vectors of the four individual learners are spliced with the standardized test feature data to construct an enhanced feature test data set:

[0061] S1055: Input the enhanced feature test data into the trained second layer TabPFN meta-learner to calculate the final prediction probability: ) S1056: Based on the preset threshold make a binary classification decision, and the internal short circuit detection result is:

[0062] wherein the threshold The optimal value is determined through but not limited to ROC curve analysis.

[0063] The battery short circuit detection method of the fusion of impedance spectrum and explainable integrated learning of the embodiment has good universality, is suitable for but not limited to different types of batteries, and can help optimize the EIS test process, significantly reduce the test time, and improve the detection efficiency by quantifying the contribution of impedance characteristics at different frequencies to the detection result to extract key impedance characteristics.

[0064] Figure 3 is a structure diagram of a battery short circuit detection system of the fusion of impedance spectrum and explainable integrated learning in the embodiment of the application, and the embodiment corresponds to the battery short circuit detection method of the fusion of impedance spectrum and explainable integrated learning of Figure 1 , as shown in Figure 3 , the battery short circuit detection system of the fusion of impedance spectrum and explainable integrated learning in the embodiment can include: A multi-dimensional impedance feature extraction module 301 is configured to obtain electrochemical impedance spectrum data of a battery at multiple frequency points, extract impedance real part and imaginary part features at different frequencies, and then obtain a standardized multi-dimensional impedance feature data set after standardization preprocessing; A first layer model training module 302 is configured to train a first layer integrated model based on the standardized multi-dimensional impedance feature data set, and obtain a prediction data set output by the first layer model; The second-layer meta-learner training module 303 is used to fuse the predicted probability dataset output by the first-layer model with the standardized multidimensional impedance feature dataset to form an enhanced feature dataset and train the second-layer meta-learner accordingly, thereby forming a two-layer ensemble model consisting of the first-layer ensemble model and the second-layer meta-learner. The key frequency band feature screening module 304 is used to perform importance analysis of local and global features on the trained two-layer ensemble model based on the SHAP interpretability analysis framework, identify the contribution of each impedance feature to internal short circuit detection at different frequencies, and screen out key frequency band features. The model retraining module 305 is used to retrain the first-layer ensemble model and the second-layer meta-learner based on important frequency band features, and to perform battery short circuit detection by retraining the optimized first-layer ensemble model and the second-layer meta-learner.

[0065] It should be noted here that, Figure 3 The fused impedance spectrum and the various modules in the interpretable ensemble learning battery short-circuit detection system, and Figure 1 The fused impedance spectrum in the method corresponds one-to-one with each step in the interpretable ensemble learning battery short-circuit detection method, and their specific implementation processes are the same, so they will not be repeated here.

[0066] Reference Figure 4 A schematic diagram of an electronic device is provided. It should be noted that... Figure 4 The electronic device 400 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0067] like Figure 4 As shown, the electronic device 400 includes a central processing unit (CPU) 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage section 408 into a random access memory (RAM) 403. The RAM 403 also stores various programs and data required for system operation. The CPU 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0068] The following components are connected to the I / O interface 405: an input section 406 including a keyboard, a mouse, etc.; an output section 407 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a local area network (LAN) card, a modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as necessary. A removable recording medium 411 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 410 as necessary, so that a computer program read out from it is installed in the storage section 408 as necessary.

[0069] The central processing unit 401 in the electronic device of the present embodiment, when executing the program, realizes the steps in the battery short circuit detection method using fusion impedance spectrum and interpretable ensemble learning as shown in Figure 1

[0070] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for executing the method shown in Figure 1 In such embodiments, the computer program can be downloaded and installed from a network via the communication section 409, and / or installed from the removable recording medium 411. When the computer program is executed by the central processing unit 401, various functions defined in the apparatus of the present application are executed.

[0071] The computer program instructions corresponding to the method shown in Figure 1 may also be stored in a computer readable storage medium that can direct the computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer readable storage medium produce a product including instruction apparatus, which realizes the functions specified in the flow Figure 1 charts or multiple flowcharts and / or blocks Figure 1 charts or multiple flowcharts and / or blocks.

[0072] It can be understood by those of ordinary skill in the art that all or part of the flowcharts in the above-mentioned embodiment methods can be implemented by a computer program instructing relevant hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it can include the flowcharts of the above-mentioned embodiment methods. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), etc. ​

[0073] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. The present application can have various changes and modifications for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A battery short-circuit detection system integrating impedance spectroscopy and interpretable integrated learning, characterized in that, include: The multidimensional impedance feature extraction module is used to acquire electrochemical impedance spectral data of the battery at multiple frequency points, extract the real and imaginary impedance features at different frequencies, and then perform standardized preprocessing to obtain a standardized multidimensional impedance feature dataset. The first-layer model training module is used to train the first-layer ensemble model based on the standardized multidimensional impedance feature dataset, and obtain the prediction dataset output by the first-layer model. The second-layer meta-learner training module is used to fuse the predicted probability dataset output by the first-layer model with the standardized multidimensional impedance feature dataset to form an enhanced feature dataset and train the second-layer meta-learner accordingly, thereby forming a two-layer ensemble model consisting of the first-layer ensemble model and the second-layer meta-learner. The key frequency band feature screening module is used to perform importance analysis of local and global features on the trained two-layer ensemble model based on the SHAP interpretability analysis framework, identify the contribution of each impedance feature to internal short circuit detection at different frequencies, and screen out key frequency band features. The model retraining module is used to retrain the first-layer ensemble model and the second-layer meta-learner based on important frequency band features, and to perform battery short circuit detection by retraining the optimized first-layer ensemble model and the second-layer meta-learner.

2. The battery short-circuit detection system integrating impedance spectroscopy and interpretable integrated learning as described in claim 1, characterized in that, Based on the SHAP interpretability analysis framework, the process of performing importance analysis on local and global features of the trained two-layer ensemble model is as follows: A SHAP interpreter is constructed based on the trained two-layer ensemble model, and the real and imaginary impedance features extracted at the original multi-frequency points are used as the interpreting objects. The contribution of the real and imaginary parts of the impedance is quantified using the Shapley value theory to obtain the Shapley value of each impedance feature. For any test sample, calculate the set of SHAP values ​​for all its impedance characteristics to obtain the local feature importance distribution of the sample; The global importance index of a certain impedance feature is obtained by averaging the absolute value of the SHAP value of each feature across all training samples.

3. The battery short-circuit detection system integrating impedance spectroscopy and interpretable integrated learning as described in claim 1 or 2, characterized in that, In the process of screening key frequency band features, the impedance features at all frequencies are globally sorted according to the value of the global importance index of a certain impedance feature, and the impedance features of the top k most important frequency points or corresponding frequency ranges are selected as key frequency band features; where k is a positive integer greater than or equal to 2.

4. The battery short-circuit detection system integrating impedance spectroscopy and interpretable integrated learning as described in claim 1, characterized in that, Based on a standardized multidimensional impedance feature dataset, a K-fold cross-validation strategy is used to train the first-layer ensemble model.

5. The battery short-circuit detection system integrating impedance spectroscopy and interpretable integrated learning as described in claim 1, characterized in that, The first-layer ensemble model includes four heterogeneous base learners: XGBoost, CatBoost, Random Forest, and TabPFN. The predicted probability output on the training set is obtained through cross-validation.

6. The battery short-circuit detection system integrating impedance spectroscopy and interpretable integrated learning as described in claim 1, characterized in that, A horizontal splicing strategy is used to fuse the predicted probability dataset output by the first-layer model with the standardized multidimensional impedance feature dataset to obtain an enhanced feature dataset.

7. The battery short-circuit detection system integrating impedance spectroscopy and interpretable integrated learning as described in claim 1, characterized in that, During the process of battery short circuit detection by retraining and optimizing the first-layer ensemble model and the second-layer meta-learner, a binary classification decision is made by comparing the probability output by the second-layer meta-learner with a preset threshold to detect whether there is a short circuit fault in the battery.

8. A battery short-circuit detection system integrating impedance spectroscopy and interpretable integrated learning, characterized in that, include: Electrochemical impedance spectroscopy data of the battery at multiple frequency points were obtained, and the real and imaginary parts of the impedance at different frequencies were extracted. After standardization preprocessing, a standardized multidimensional impedance feature dataset was obtained. Based on the standardized multidimensional impedance feature dataset, the first-layer ensemble model is trained to obtain the prediction dataset output by the first-layer model; The predicted probability dataset output by the first-layer model is fused with the standardized multidimensional impedance feature dataset to form an enhanced feature dataset, which is then used to train the second-layer meta-learner, thus forming a two-layer ensemble model consisting of the first-layer ensemble model and the second-layer meta-learner. Based on the SHAP interpretability analysis framework, the importance of local and global features of the trained two-layer ensemble model is analyzed to identify the contribution of each impedance feature to internal short circuit detection at different frequencies and to screen out key frequency band features. Based on the features of important frequency bands, the first-layer ensemble model and the second-layer meta-learner are retrained, and battery short circuit detection is performed by retraining the optimized first-layer ensemble model and the second-layer meta-learner.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the steps in the battery short-circuit detection method based on fused impedance spectrum and interpretable integrated learning as described in claim 8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the battery short-circuit detection method based on fused impedance spectrum and interpretable integrated learning as described in claim 8.

Images