Lithium ion battery internal short circuit detection method based on high-dimensional frequency response mapping
By constructing the high-dimensional frequency response characteristics of lithium-ion batteries, and using random forest and core principal component analysis for feature screening and dimensionality reduction, combined with deep neural network models, the problem of insufficient short-circuit detection accuracy and real-time performance in lithium-ion batteries in the existing technology is solved, achieving higher detection accuracy and model generalization capabilities.
Patent Information
- Application Number
- CN202411858357.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-04-08
AI Technical Summary
In the prior art, when detecting short circuit failures in lithium-ion batteries, the accuracy and real-time performance are insufficient, and early signs cannot be captured in time. The electrochemical impedance spectroscopy method based on a single or a few frequency points has problems such as limited feature expression capabilities and a lot of redundant information.
Using a method based on correlation coefficient, high-dimensional frequency response characteristics are constructed by collecting electrochemical impedance spectral data of multi-frequency points, flattening processing and introducing higher-order quadratic terms. Then, random forest and nuclear principal component analysis were used for feature screening and dimensionality reduction, redundant features were removed, and important features related to internal short circuits were retained. Finally, the deredundant features are trained using the deep neural network model to generate an internal short circuit detection model.
It improves the sensitivity and accuracy of short-circuit detection in lithium-ion batteries, reduces the feature dimension, improves the generalization ability of the model, and is suitable for real-time monitoring and safety warning of lithium-ion batteries.
Smart Images
Figure CN119780722A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of lithium-ion battery safety monitoring, and in particular relates to a method for detecting short circuits in lithium-ion batteries. The method aims to improve the detection accuracy and stability of short-circuit faults in batteries through the high-dimensional frequency response characteristics of electrochemical impedance data at multiple frequency points. The method is applicable to various types of lithium-ion batteries and is widely used in scenarios requiring high safety and reliability, such as electric vehicles, power storage systems, and consumer electronics. Background Art
[0002] Lithium-ion batteries have been widely used in electric vehicles, portable electronic devices and energy storage systems due to their high energy density and long service life. However, lithium-ion batteries are prone to internal short circuit failures under high-frequency charging and discharging or excessive use, causing thermal runaway and posing serious safety hazards. Traditional internal short circuit detection methods mainly rely on temperature monitoring, voltage and current analysis, etc., but these methods have certain limitations in detection accuracy and real-time performance, and cannot capture early signs of internal short circuits in a timely manner.
[0003] Electrochemical impedance spectroscopy (EIS) is a technology widely used in battery health monitoring, which can reflect changes in the internal state of the battery through frequency response characteristics. However, traditional EIS methods mostly rely on the impedance characteristics of a single or a few frequency points, ignoring the dynamic relationship between multiple frequency responses, resulting in limited feature expression capabilities of the detection model. At the same time, frequency response data with higher feature dimensions often contain more redundant information, which increases the training complexity of the model and reduces its generalization ability.
[0004] In recent years, with the application of machine learning and deep learning in the field of fault detection, some detection methods based on deep neural networks have gradually emerged. Although these methods have improved the detection accuracy to a certain extent, they still face problems such as insufficient feature extraction, excessive complexity of model structure, and large demand for computing resources. Therefore, there is an urgent need for a new detection method that can effectively capture the early characteristics of short circuits in lithium-ion batteries based on multi-frequency response characteristics, and has high computational efficiency and strong generalization ability. The lithium-ion battery internal short circuit detection method based on high-dimensional frequency response mapping provided by the present invention constructs high-dimensional dynamic features of multiple frequency points, and removes data redundancy through feature screening and dimensionality reduction processing, which can effectively improve the performance and generalization ability of the detection model, and has important practical value and application prospects. Summary of the invention
[0005] In view of this, the present invention provides a high-dimensional frequency response mapping method based on lithium-ion battery internal short circuit detection, which realizes rapid online detection while effectively improving the early detection capability of lithium-ion battery internal short circuit.
[0006] The technical solution adopted by the present invention is:
[0007] A method for detecting internal short circuit of a lithium-ion battery based on correlation coefficient comprises the following steps:
[0008] S1: Collect electrochemical impedance spectroscopy (EIS) data of lithium-ion batteries at different frequency points;
[0009] S2: Flatten the impedance spectrum characteristics to generate high-dimensional frequency response characteristics at multiple frequency points;
[0010] S3: Through random forest feature screening and kernel principal component analysis (Kernel PCA) dimensionality reduction, redundant features are removed from the generated high-dimensional frequency response features, and important features related to internal short circuits are retained;
[0011] S4: Use a deep neural network model to train the high-dimensional frequency response features after redundancy removal to generate the final internal short circuit detection model;
[0012] Furthermore, the step S1 is specifically as follows: injecting a small disturbance signal into the battery in a normal working state, collecting the response data on both sides of the current and voltage, and performing frequency domain transformation on the obtained data to respectively extract basic features such as the real part of impedance, the imaginary part of impedance, the impedance modulus, and the impedance phase.
[0013] Furthermore, the step S2 is specifically as follows: after obtaining the basic features of the response data, on this basis, flattening the impedance spectrum features to generate high-dimensional frequency response features at multiple frequency points, and introducing high-order quadratic terms to retain more complex nonlinear relationships between the features.
[0014] Further, the step S3 is specifically as follows: by random forest feature screening and kernel principal component analysis (KernelPCA) dimensionality reduction, redundant features are removed from the generated high-dimensional frequency response features, and important features related to the internal short circuit are retained. The formula for screening the optimal features using the random forest algorithm is:
[0015] Where, is the number of decision trees in the forest. is each decision tree, where represents the th tree.
[0016] The 10 best features are mapped to a high-dimensional space. The mapping function is expressed as . The similarity between samples is calculated by the kernel function. At the same time, in order to avoid offset, the kernel matrix needs to be centered. The calculation formula is:
[0017] Among them, 1 represents the average matrix of the unit matrix. On the centralized kernel matrix, the eigenvalue and eigenvector are solved, and the calculation formula is:
[0018] The first k eigenvectors with larger eigenvalues are retained to obtain the feature representation after dimensionality reduction.
[0019] Finally, the low-dimensional features are obtained. The projection of each new sample point in the reduced-dimensional space is calculated by the following formula:
[0020] Among them, is the component of the feature vector, and is the original 10 filtered features.
[0021] Furthermore, the step S4 is specifically as follows: using a deep neural network model to train the high-dimensional frequency response features after removing redundancy to generate a final internal short circuit detection model.
[0022] (1) In the input layer, DNN receives the low-dimensional feature vector after KPCA dimension reduction as input. This feature vector retains the key information related to the detection of short circuits in batteries and removes redundancy and noise to improve the training efficiency and generalization performance of the model. The input format of each sample in DNN is a feature vector, which is represented as.
[0023] (2) In the hidden layer, the first layer receives data from the input layer and performs a nonlinear transformation on the input through the activation function to capture the nonlinear relationship between features. The second layer processes the output features of the first layer, further refines the feature information, gradually aggregates the high-dimensional information in the feature space, reduces the interference of redundant features, and gradually converges. The third layer further aggregates and refines the feature information to form the final high-level feature representation, and performs anti-overfitting processing.
[0024] (3) Finally, the output layer of the DNN is a single neuron with an output value between 0 and 1, which is used for binary classification tasks. In the end, "0" indicates that the battery is in a normal state, and "1" indicates that the battery has an internal short circuit fault.
[0025] The beneficial effect of the present invention is that the present invention constructs the dynamic characteristics of multiple frequency points of lithium-ion batteries through high-dimensional frequency response mapping, which effectively enhances the sensitivity of internal short circuit detection. At the same time, through random forest and kernel principal component analysis to remove redundancy, the feature dimension is reduced and the generalization ability of the model is improved. Experimental results show that the method of the present invention is significantly superior to traditional methods in terms of accuracy and computational efficiency, and is suitable for real-time monitoring and safety warning of lithium-ion batteries. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 :Flowchart of short circuit detection in lithium-ion batteries based on high-dimensional frequency response mapping DETAILED DESCRIPTION
[0027] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0028] The present invention provides a high-dimensional frequency response mapping method based on lithium-ion battery internal short circuit detection, and the implementation steps include:
[0029] S1: Collect electrochemical impedance spectroscopy (EIS) data of lithium-ion batteries at different frequency points. The specific implementation method is as follows:
[0030] (1) Under normal constant current charging conditions, inject a small amplitude disturbance signal;
[0031] (2) Collect current and voltage response data on both sides of the battery;
[0032] (3) Perform frequency domain transformation on the collected current and voltage data to calculate the impedance spectrum data of the battery.
[0033] (4) Extract the real part of impedance, the imaginary part of impedance, and the impedance modulus from the obtained impedance spectrum data.
[0034] Impedance modulus.
[0035] S2: Flatten the impedance spectrum features to generate high-dimensional frequency response features at multiple frequency points. And introduce high-order quadratic terms to retain more complex nonlinear relationships between its features.
[0036] S3: Through random forest feature screening and kernel principal component analysis (Kernel PCA) dimensionality reduction, redundant features are removed from the generated high-dimensional frequency response features, retaining important features related to internal short circuits:
[0037]
[0038] Where B is the number of decision trees in the forest. b is each decision tree, where b represents the bth tree.
[0039] The 10 best features obtained Mapped to high-dimensional space, the mapping function is expressed as Through the kernel function Calculate the similarity between samples. At the same time, in order to avoid deviation, the kernel matrix K needs to be centralized, and its calculation formula is:
[0040] K'=K-1K-K1+1K1
[0041] Among them, 1 represents the average matrix of the unit matrix. On the centralized kernel matrix K', the eigenvalue and eigenvector are solved, and the calculation formula is:
[0042] K'v=λv
[0043] The first k eigenvectors with larger eigenvalues are retained to obtain the feature representation after dimensionality reduction.
[0044] Finally, the low-dimensional features are obtained. The projection of each new sample point y in the reduced-dimensional space is calculated by the following formula:
[0045]
[0046] Among them, α i are the components of the eigenvector, are the original 10 filtered features.
[0047] S4: Use the deep neural network model to train the high-dimensional frequency response features after redundancy removal to generate the final internal short circuit detection model:
[0048] (1) In the input layer, DNN receives the low-dimensional feature vector after KPCA dimensionality reduction as input. This feature vector retains the key information related to the detection of short circuits in batteries and removes redundancy and noise to improve the training efficiency and generalization performance of the model. The input format of each sample in DNN is a feature vector, which is represented as [f1,f2,…,f 10 ].
[0049] (2) In the hidden layer, the first layer receives data from the input layer and performs a nonlinear transformation on the input through the activation function to capture the nonlinear relationship between features. The second layer processes the output features of the first layer, further refines the feature information, gradually aggregates the high-dimensional information in the feature space, reduces the interference of redundant features, and gradually converges. The third layer further aggregates and refines the feature information to form the final high-level feature representation, and performs anti-overfitting processing.
[0050] (3) Finally, the output layer of the DNN is a single neuron with an output value between 0 and 1, which is used for binary classification tasks. In the end, "0" indicates that the battery is in a normal state, and "1" indicates that the battery has an internal short circuit fault.
Claims
1. A method for detecting internal short circuits in lithium-ion batteries based on high-dimensional frequency response mapping, characterized in that: The following steps are involved: S1: Collect electrochemical impedance spectroscopy (EIS) data of lithium-ion batteries at different frequency points; S2: Flatten the impedance spectrum characteristics to generate high-dimensional frequency response characteristics at multiple frequency points; S3: Through random forest feature screening and kernel principal component analysis (Kernel PCA) dimensionality reduction, redundant features are removed from the generated high-dimensional frequency response features, and important features related to internal short circuits are retained; S4: Use a deep neural network model to train the high-dimensional frequency response features after removing redundancy to generate the final internal short circuit detection model.
2. The method for detecting internal short circuit of a lithium-ion battery based on high-dimensional frequency response mapping according to claim 1, characterized in that: In step S1, EIS data of lithium-ion batteries at different frequency points are collected: (1) Under normal constant current charging conditions, inject a small amplitude disturbance signal; (2) Collect current and voltage response data on both sides of the battery; (3) Performing frequency domain transformation on the collected current and voltage data to calculate the impedance spectrum data of the battery; (4) Based on the obtained impedance spectrum data, the real part of impedance, the imaginary part of impedance, the impedance modulus, and the impedance modulus are extracted respectively.
3. A lithium-ion battery internal short circuit detection method based on high-dimensional frequency response mapping as claimed in claim 1, characterized in that: After acquiring the impedance spectrum data in step S2, the impedance spectrum features are flattened to generate high-dimensional frequency response features at multiple frequency points, and high-order quadratic terms are introduced to retain more complex nonlinear relationships between the features.
4. A lithium-ion battery internal short circuit detection method based on high-dimensional frequency response mapping as claimed in claim 1, characterized in that: In step S3, the obtained high-dimensional frequency response features are subjected to dimensionality reduction through random forest feature screening and kernel principal component analysis (Kernel PCA).
5. A lithium-ion battery internal short circuit detection method based on high-dimensional frequency response mapping as claimed in claim 1, characterized in that: In step S4, a deep neural network is used to train the high-dimensional frequency response features after redundancy removal to generate a final internal short circuit detection model.