Energy storage battery health state strengthening diagnosis method based on battery body characteristic analysis
By constructing a multimodal mathematical model based on the battery's intrinsic characteristics, and combining ultrasonic imaging and electrochemical impedance data, a hybrid neural network is used to diagnose the battery's health status, solving the problem of insufficient diagnostic accuracy in existing technologies and realizing early fault warning and full life cycle management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XINXIANG POWER SUPPLY COMPANY STATE GRID HENAN ELECTRIC POWER
- Filing Date
- 2025-12-19
- Publication Date
- 2026-05-05
AI Technical Summary
Existing methods for diagnosing the health status of energy storage batteries rely on external parameters and cannot monitor the internal state of the battery in real time. Furthermore, traditional electrochemical methods cannot capture complex physicochemical processes, resulting in insufficient diagnostic accuracy and limited dimensionality.
A multimodal mathematical model based on battery body feature analysis is constructed. Combined with ultrasonic imaging, material composition and electrochemical impedance data, a hybrid neural network of bidirectional long short-term memory network and convolutional neural network is used for training to output battery health status diagnosis results.
It significantly improves diagnostic accuracy, enables early fault warning, supports full lifecycle management, is applicable to various battery types, and can be integrated into existing battery management systems.
Smart Images
Figure CN121978567A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of energy storage battery health status diagnosis technology, specifically relating to an enhanced diagnosis method for energy storage battery health status based on battery body feature analysis. Background Technology
[0002] Current state of health (SOH) diagnosis of energy storage batteries mainly relies on external parameters (such as voltage, current, and temperature) and traditional electrochemical methods (such as electrochemical impedance spectroscopy, EIS). However, these methods have significant drawbacks:
[0003] 1. Limitations of external parameters: Signals such as voltage and current are easily affected by operating conditions and cannot directly reflect the battery's internal characteristics such as material degradation and electrolyte drying.
[0004] 2. Shortcomings of traditional electrochemical methods: EIS requires offline testing and cannot be monitored in real time; disassembly analysis is a destructive test and cannot be used for batteries in operation.
[0005] 3. Insufficient model accuracy: Existing data-driven models (such as support vector machines) rely solely on external parameters, making it difficult to capture the complex physicochemical processes inside batteries.
[0006] Currently, although the ultrasonic scanning imaging technology developed by the team at Huazhong University of Science and Technology can detect the state of gas generation and electrolyte wetting inside the battery in real time, it has not been deeply integrated with other physical characteristics (such as material composition and microstructure) and mathematical models, resulting in a single diagnostic dimension. Summary of the Invention
[0007] To address the shortcomings of existing technologies, this invention aims to provide an enhanced diagnostic method for the health status of energy storage batteries based on battery intrinsic feature analysis. This method constructs a multimodal mathematical model based on battery intrinsic feature analysis, breaking through the dependence of traditional methods on external parameters and enabling enhanced diagnostics of the health status of energy storage batteries.
[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0009] A method for enhanced diagnosis of the health status of energy storage batteries based on battery intrinsic feature analysis includes the following steps:
[0010] S1. Collect battery body characteristic data, including ultrasonic imaging data, material composition data and electrochemical impedance data;
[0011] S2. Preprocess and extract features from the feature data:
[0012] S3. Construct a hybrid neural network model based on a bidirectional long short-term memory network and a convolutional neural network, and train it by inputting the feature data;
[0013] S4. Use the trained model to output battery health status diagnosis results, including the comprehensive health index and each sub-indicator.
[0014] Furthermore, the ultrasonic imaging data in step S1 is acquired using high-frequency ultrasonic transmission technology, including sub-millimeter-level imaging data such as the electrolyte wetting state and gas production distribution inside the battery.
[0015] Material composition data, including electrode material lattice parameters, particle size, and elemental content, are obtained through methods including X-ray diffraction and scanning electron microscopy.
[0016] Electrochemical impedance data, including charge transfer resistance and ion diffusion coefficient, were analyzed using a fractional-order equivalent circuit model.
[0017] Furthermore, in step S1, the ultrasonic detection during ultrasonic imaging data acquisition involves using a high-frequency ultrasonic probe to scan the battery, generating a grayscale image of the internal structure through a signal processing algorithm, and extracting features such as the number of bubbles and the electrolyte wetting area.
[0018] Ultrasonic imaging data is obtained by scanning the battery with a high-frequency ultrasonic probe and receiving the echo signal U(t). After signal processing, the grayscale image I(x,y) of the internal structure is reconstructed using a back projection algorithm or a delay superposition algorithm, where (x,y) are the image pixel coordinates.
[0019] Material analysis during material composition data acquisition: Battery electrode samples are collected periodically, crystal structure changes are analyzed by XRD, particle morphology evolution is observed by SEM, and a material degradation database is established;
[0020] Material composition data were obtained by X-ray diffraction (XRD) to obtain diffraction patterns. The lattice spacing d was calculated using the Bragg equation: nλ = 2dsinθ, where λ is the X-ray wavelength, θ is the diffraction angle, and n is the diffraction order.
[0021] EIS testing during electrochemical impedance data acquisition: EIS tests were performed under different SOC states, and charge transfer resistance and Warburg impedance parameters were obtained by fitting the FECM model.
[0022] Electrochemical impedance data were measured by EIS using an electrochemical workstation to obtain complex impedance spectra: Z(ω) = Z'(ω) + jZ''(ω);
[0023] Where ω is the angular frequency, Z'(ω) is the real part, Z''(ω) is the imaginary part, and j is the imaginary unit (i.e., ji). 2 =-1); The fractional-order equivalent circuit model FECM is used for fitting, and the model expression is: ;
[0024] in, For equivalent series resistance, For charge transfer resistor, It is a constant phase angle element. Its fractional exponent, This is the Warburg coefficient.
[0025] Furthermore, in step S2, feature preprocessing requires standardization and noise reduction of the ultrasound images, material composition data, and EIS parameters;
[0026] Feature engineering includes:
[0027] Ultrasonic characteristics: bubble density, wetting rate, sound velocity attenuation coefficient;
[0028] Material characteristics: lattice constant, particle size distribution, elemental content;
[0029] EIS characteristics: R ct Zw, equivalent series resistance.
[0030] Step S2 specifically includes:
[0031] S2-1. Data Standardization: Perform Z-score standardization on data from different sources with different dimensions;
[0032] S2-2. Ultrasonic feature extraction, including:
[0033] bubble density : ;
[0034] in, This represents the total number of bubble outlines identified by the image segmentation algorithm. The area of the region of interest (ROI);
[0035] wetting rate : ;
[0036] in, The area of the electrolyte-wetting region is distinguished by a grayscale threshold. This represents the total area of the electrode region;
[0037] S2-3. Material Feature Extraction: Lattice constants (a, c) are obtained from XRD patterns through peak position fitting; average particle size is obtained from SEM images through image analysis. and its distribution standard deviation ;
[0038] S2-4. EIS Feature Extraction: Obtain parameter values directly from the fitted FECM model.
[0039] Furthermore, in step S3, the hybrid neural network model employs a genetic algorithm to optimize hyperparameters; including:
[0040] CNN layer: used to perform convolution operations on ultrasound images to extract spatial features, including defect locations and structural anomalies;
[0041] BiLSTM layer: used to process temporal features, including material degradation trends and EIS parameter changes, in order to capture long-term dependencies.
[0042] Fully connected layer: used to fuse multimodal features and output CHI and various sub-indicators, including SOH and thermal runaway risk.
[0043] Furthermore, the hybrid neural network model training strategy in step S3 includes:
[0044] Dataset: Collect battery samples at different aging stages and label the actual SOH values;
[0045] Loss function: Joint optimization of root mean square error (RMSE) and cross-entropy loss;
[0046] Optimizer: Adagrad algorithm, with adaptive learning rate adjustment.
[0047] The hybrid neural network model is constructed in step S3 as follows:
[0048] CNN section: Let the input ultrasound image be... After the ll-th convolutional layer: ;
[0049] in, Indicates the first The feature map output by the convolution operation is the result after convolution, biasing, and ReLU activation. Indicates the first Input feature map for layer convolution operation. , The weights and biases of the convolutional kernel in this layer are... This represents the convolution operation. The activation function is used; finally, the image feature vector is obtained through global average pooling. ;
[0050] BiLSTM section: Processing time-series materials and EIS feature sequences: ;
[0051] in, Let t be the material characteristic vector (such as lattice constant, particle size, etc.) at time t. The eigenvectors of the electrochemical impedance spectrum at time t (such as charge transfer resistance, Warburg impedance, etc.) together constitute the time-series input sequence.
[0052] Forward LSTM computes hidden state Backward LSTM computation The final output is obtained by splicing the data at each time step. Finally, the output of the last time step is taken or temporal pooling is performed to obtain the temporal feature vector. ;
[0053] Feature fusion and output: and Concatenate, input fully connected layer: ;
[0054] in, , These are the parameters for the fully connected layer. It is a linear activation function, and its output is a health status indicator;
[0055] Hyperparameter optimization: The learning rate, number of network layers, and number of neurons are optimized using the genetic algorithm (GA). The fitness function is defined as the reciprocal of the model's diagnostic error on the validation set. The optimal combination of hyperparameters is searched iteratively through selection, crossover, and mutation operations.
[0056] Furthermore, the diagnostic results in step S4 also include quantitative indicators of battery health status and dynamic threshold warnings; specifically including:
[0057] The Comprehensive Health Index (CHI), as the main output of the model, can be defined as: ;
[0058] in, For sub-indicators of health status, the model directly outputs the values; For the normalized thermal runaway risk score, For consistency scoring functions of other features, , , The weighting coefficients and ;
[0059] Dynamic threshold warning: Warning threshold T alarm It can be dynamically adjusted according to the working conditions: ;
[0060] in, Based on the threshold, , For temperature and charge / discharge rate Adjustment coefficient.
[0061] The beneficial effects of this invention are as follows:
[0062] 1. This invention can significantly improve diagnostic accuracy: by combining ultrasound, materials, and EIS multi-dimensional features, the SOH diagnostic error is reduced to within ±2%, which is better than the traditional method (error ±5%).
[0063] 2. This invention enables early fault warning: potential problems such as electrolyte drying can be detected up to 300 cycles in advance through ultrasonic imaging.
[0064] 3. This invention enables full lifecycle management: it supports real-time monitoring of battery health status and lifespan prediction, extending system service time by 15%-20%.
[0065] 4. This invention significantly improves compatibility and scalability: it is applicable to various battery types such as lithium-ion and vanadium redox flow batteries, and can be integrated into existing battery management systems (BMS). Attached Figure Description
[0066] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation
[0067] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention.
[0068] This invention proposes an enhanced diagnostic method for the health status of energy storage batteries based on battery intrinsic feature analysis, comprising the following steps:
[0069] S1. Collect battery body characteristic data, ultrasonic imaging data, material composition data, and electrochemical impedance data. Specific details are as follows:
[0070] Ultrasonic imaging data is obtained by scanning the battery with a high-frequency ultrasonic probe and receiving the echo signal U(t). After signal processing (such as filtering and gain adjustment), the internal structure grayscale image I(x,y) is reconstructed using a back-projection algorithm or a delay-overlay algorithm, where (x,y) are the image pixel coordinates.
[0071] Material composition data were analyzed to obtain diffraction patterns using X-ray diffraction (XRD). The lattice spacing d was calculated using the Bragg equation nλ = 2dsinθ. Here, λ is the X-ray wavelength, θ is the diffraction angle, and n is the diffraction order.
[0072] Electrochemical impedance data were measured by EIS using an electrochemical workstation to obtain complex impedance spectra: Z(ω) = Z'(ω) + jZ''(ω);
[0073] Where ω is the angular frequency, Z'(ω) is the real part, Z''(ω) is the imaginary part, and j is the imaginary unit (i.e., j...2 =-1). A fractional-order equivalent circuit model (FECM) is used for fitting; the model expression is: ;
[0074] in, For equivalent series resistance, For charge transfer resistor, It is a constant phase angle element. Its fractional exponent, This is the Warburg coefficient.
[0075] S2. Preprocess and extract features from the feature data. This includes:
[0076] Feature preprocessing requires standardization and noise reduction of ultrasound images, material composition data, and EIS parameters;
[0077] Feature engineering includes:
[0078] Ultrasonic characteristics: bubble density, wetting rate, sound velocity attenuation coefficient;
[0079] Material characteristics: lattice constant, particle size distribution, elemental content;
[0080] EIS characteristics: R ct Zw, equivalent series resistance.
[0081] Specifically: (1) Data standardization: Z-score standardization is performed on data from different sources with different dimensions.
[0082] (2) Ultrasound image feature extraction:
[0083] bubble density ( ): ;
[0084] in, This represents the total number of bubble outlines identified using image segmentation algorithms (such as the Otsu thresholding method). The area of the region of interest (ROI).
[0085] wetting rate ( ): ;
[0086] in, The area of the electrolyte-wetted region (distinguished by grayscale threshold). This represents the total area of the electrode region.
[0087] Material feature extraction: Lattice constants (a, c) were obtained from XRD patterns by peak fitting; average particle size was obtained from SEM images by image analysis. ) and its distribution standard deviation .
[0088] EIS feature extraction: Parameter values are obtained directly from the fitted FECM model.
[0089] S3. Construct a hybrid neural model based on a bidirectional long short-term memory network and a convolutional neural network, and train it using the input feature data. The model is constructed as follows:
[0090] CNN section: Let the input ultrasound image be... After the ll-th convolutional layer: ;
[0091] in, Indicates the first The feature map output by the convolution operation is the result after convolution, biasing, and ReLU activation. Indicates the first Input feature map for layer convolution operation. , The kernel weights and biases of this layer This represents the convolution operation. The activation function is used. Finally, global average pooling is used to obtain the image feature vector. .
[0092] BiLSTM section: Processing time-series materials and EIS feature sequences: ;
[0093] in, Let t be the material characteristic vector (such as lattice constant, particle size, etc.) at time t. The eigenvectors of the electrochemical impedance spectrum at time t (such as charge transfer resistance, Warburg impedance, etc.) together constitute the time-series input sequence.
[0094] Forward LSTM computes hidden state Backward LSTM computation The final output is obtained by splicing the data at each time step. Finally, the output of the last time step is taken or temporal pooling is performed to obtain the temporal feature vector. .
[0095] Feature fusion and output: and Concatenate, input fully connected layer: ;
[0096] in, , These are the parameters for the fully connected layer. It is a linear activation function, and its output is a health status indicator.
[0097] Hyperparameter optimization: Genetic algorithm (GA) is used to optimize hyperparameters such as learning rate, number of network layers, and number of neurons. The fitness function is defined as the reciprocal of the model's diagnostic error on the validation set. The optimal combination of hyperparameters is searched iteratively through selection, crossover, and mutation operations.
[0098] like Figure 1 As shown, the hybrid neural network model includes:
[0099] CNN layer: used to perform convolution operations on ultrasound images to extract spatial features, including defect locations and structural anomalies;
[0100] BiLSTM layer: used to process temporal features, including material degradation trends and EIS parameter changes, in order to capture long-term dependencies.
[0101] Fully connected layer: used to fuse multimodal features and output CHI and various sub-indicators, including SOH and thermal runaway risk.
[0102] Hybrid neural network model training strategies include:
[0103] Dataset: Collect battery samples at different aging stages and label the actual SOH values;
[0104] Loss function: Joint optimization of root mean square error (RMSE) and cross-entropy loss;
[0105] Optimizer: Adagrad algorithm, with adaptive learning rate adjustment.
[0106] S4. Utilize the trained model to output battery health status diagnostic results, including comprehensive health index and its sub-indicators, quantitative indicators of battery health status, and dynamic threshold warnings.
[0107] Comprehensive Health Index (CHI): As the main output of the model, it can be defined as: ;
[0108] in, This is a sub-indicator of health status (values directly output by the model). For the normalized thermal runaway risk score, For consistency scoring functions of other features, , , The weighting coefficients and .
[0109] Dynamic threshold warning: The warning threshold Talarm can be dynamically adjusted according to operating conditions. ;
[0110] in, Based on the threshold, , For temperature and charge / discharge rate Adjustment coefficient.
[0111] Health status diagnosis and early warning have the following advantages:
[0112] Real-time diagnostics: Input the real-time collected ontology features into the trained model and output CHI and its sub-indices.
[0113] Dynamic threshold setting: The diagnostic thresholds are dynamically adjusted based on operating conditions such as battery type, ambient temperature, and charge / discharge rate. For example, the warning threshold for thermal runaway risk is lowered under high-temperature environments.
[0114] Visual interface: Displays battery health status through a graphical interface, marks abnormal areas (such as areas where the electrolyte has dried out), and provides lifespan prediction curves.
[0115] Based on the aforementioned enhanced diagnostic method for the health status of energy storage batteries based on battery body feature analysis, this invention also proposes an enhanced diagnostic system for the health status of energy storage batteries, including a body feature acquisition module, a feature processing module, a hybrid neural network model, and a visualization and early warning module. The body feature acquisition module is used to acquire ultrasonic imaging data, material composition data, and electrochemical impedance data; the feature processing module is used to preprocess and extract features from the feature data; the hybrid neural network model is used to output the battery health status diagnostic results; and the visualization and early warning module is used to display the diagnostic results and issue dynamic threshold early warnings.
[0116] Simulation experiment:
[0117] Example 1: An example of using the enhanced health status diagnosis method for energy storage batteries based on battery body feature analysis proposed in this invention for lithium-ion battery health diagnosis:
[0118] 1. Feature Acquisition:
[0119] Ultrasonic testing: An ultrasonic scan of a 280Ah lithium-ion battery revealed a cluster of bubbles with a diameter of approximately 0.5mm in the positive electrode region.
[0120] Materials analysis: XRD showed that the lattice parameter of the cathode material LiCoO2 increased by 0.3%, and SEM observed cracks on the particle surface.
[0121] EIS test: Rct increased from the initial 20mΩ to 35mΩ, and Zw increased by 40%.
[0122] 2. Model Input:
[0123] Ultrasonic characteristics: bubble density (0.8 bubbles / mm²), infiltration rate (82%).
[0124] Material characteristics: lattice constant (2.81 Å), particle size (10-15 μm).
[0125] EIS characteristics: Rct (35mΩ), Zw (0.15Ω·s^0.5).
[0126] 3. Diagnostic results:
[0127] CHI: 78% (Good health, but attention should be paid to bubble growth).
[0128] SOH: 85% (capacity decay 15%).
[0129] Thermal runaway risk: Low (no obvious signs of lithium plating).
[0130] Example 2: Using the enhanced health status diagnosis method for energy storage batteries based on battery body feature analysis proposed in this invention for vanadium redox flow battery health diagnosis:
[0131] 1. Feature Acquisition:
[0132] Ultrasonic testing revealed localized blockage in the electrolyte channel, with an ultrasonic signal attenuation rate of 15%.
[0133] Materials analysis: The thickness of the proton exchange membrane decreased by 10%, and micropores appeared on the surface.
[0134] EIS test: ESR increased from 0.5Ω to 0.8Ω, indicating a significant increase in charge transfer resistance.
[0135] 2. Model Input:
[0136] Ultrasonic features: channel blockage area (5 mm²), signal attenuation rate (15%).
[0137] Material characteristics: membrane thickness (50 μm), micropore density (2 pores / mm²).
[0138] EIS characteristics: ESR (0.8Ω), Rct (1.2Ω).
[0139] 3. Diagnostic results:
[0140] CHI: 65% (moderate health condition, electrolyte circulation system needs to be checked).
[0141] SOH: 72% (capacity decay 28%).
[0142] Thermal runaway risk: Medium (local overheating of electrolyte).
[0143] This method constructs a multimodal mathematical model based on the analysis of battery characteristics, breaking through the dependence of traditional methods on external parameters. It can achieve enhanced diagnosis of the health status of energy storage batteries, significantly improve diagnostic accuracy, realize early fault warning and full life cycle management, and significantly improve compatibility and scalability.
[0144] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.
Claims
1. A method for enhanced diagnosis of the health status of energy storage batteries based on battery intrinsic feature analysis, characterized in that, Includes the following steps: S1. Collect battery body characteristic data, including ultrasonic imaging data, material composition data and electrochemical impedance data; S2. Preprocess and extract features from the feature data: S3. Construct a hybrid neural network model based on a bidirectional long short-term memory network and a convolutional neural network, and train it by inputting the feature data; S4. Use the trained model to output battery health status diagnosis results, including the comprehensive health index and each sub-indicator.
2. The enhanced diagnostic method for the health status of energy storage batteries based on battery body feature analysis according to claim 1, characterized in that: The ultrasonic imaging data in step S1 is obtained using high-frequency ultrasonic transmission technology, including sub-millimeter-level imaging data such as the electrolyte wetting state and gas production distribution inside the battery. Material composition data, including electrode material lattice parameters, particle size, and elemental content, are obtained through methods including X-ray diffraction and scanning electron microscopy. Electrochemical impedance data, including charge transfer resistance and ion diffusion coefficient, were analyzed using a fractional-order equivalent circuit model.
3. The method for enhanced diagnosis of energy storage battery health status based on battery body feature analysis according to claim 2, characterized in that: Ultrasonic detection during ultrasound imaging data acquisition in step S1: A high-frequency ultrasonic probe was used to scan the battery, and a grayscale image of the internal structure was generated by a signal processing algorithm to extract features such as the number of bubbles and the electrolyte wetting area. Ultrasonic imaging data is obtained by scanning the battery with a high-frequency ultrasonic probe and receiving the echo signal U(t). After signal processing, the grayscale image I(x,y) of the internal structure is reconstructed using a back projection algorithm or a delay superposition algorithm, where (x,y) are the image pixel coordinates. Material analysis during material composition data acquisition: Battery electrode samples are collected regularly, crystal structure changes are analyzed by XRD, particle morphology evolution is observed by SEM, and a material degradation database is established. Material composition data were obtained by X-ray diffraction (XRD) to obtain diffraction patterns. The lattice spacing d was calculated using the Bragg equation: nλ = 2dsinθ, where λ is the X-ray wavelength, θ is the diffraction angle, and n is the diffraction order. EIS testing during electrochemical impedance spectroscopy data acquisition: EIS tests were performed under different SOC conditions, and the charge transfer resistance and Warburg impedance parameters were obtained by fitting the FECM model. Electrochemical impedance data were measured by EIS using an electrochemical workstation to obtain complex impedance spectra: Z(ω) = Z'(ω) + jZ''(ω); Where ω is the angular frequency, Z'(ω) is the real part, Z''(ω) is the imaginary part, and j is the imaginary unit. 2 =-1; Fitting is performed using the fractional-order equivalent circuit model FECM, model expression: ; in, For equivalent series resistance, For charge transfer resistor, It is a constant phase angle element. Its fractional exponent, This is the Warburg coefficient.
4. The enhanced diagnostic method for the health status of energy storage batteries based on battery body feature analysis according to claim 1, characterized in that: In step S2, feature preprocessing requires standardization and noise reduction of the ultrasound images, material composition data, and EIS parameters. Feature engineering includes: Ultrasonic characteristics: bubble density, wetting rate, sound velocity attenuation coefficient; Material characteristics: lattice constant, particle size distribution, elemental content; EIS characteristics: R ct Zw, equivalent series resistance.
5. The method for enhanced diagnosis of energy storage battery health status based on battery body feature analysis according to claim 4, characterized in that: Step S2 specifically includes: S2-1. Data Standardization: Perform Z-score standardization on data from different sources with different dimensions; S2-2. Ultrasonic feature extraction, including: bubble density : ; in, This represents the total number of bubble outlines identified by the image segmentation algorithm. The area of the region of interest (ROI); wetting rate : ; in, The area of the electrolyte-wetting region is distinguished by a grayscale threshold. This represents the total area of the electrode region; S2-3. Material Feature Extraction: Lattice constants (a, c) are obtained from XRD patterns through peak position fitting; average particle size is obtained from SEM images through image analysis. and its distribution standard deviation ; S2-4. EIS Feature Extraction: Obtain parameter values directly from the fitted FECM model.
6. The method for enhanced diagnosis of energy storage battery health status based on battery body feature analysis according to claim 1, characterized in that: In step S3, the hybrid neural network model uses a genetic algorithm to optimize hyperparameters; this hybrid neural network model includes: CNN layer: used to perform convolution operations on ultrasound images to extract spatial features, including defect locations and structural anomalies; BiLSTM layer: used to process temporal features, including material degradation trends and EIS parameter changes, in order to capture long-term dependencies; Fully connected layer: used to fuse multimodal features and output CHI and various sub-indicators, including SOH and thermal runaway risk.
7. The enhanced diagnostic method for the health status of energy storage batteries based on battery body feature analysis according to claim 6, characterized in that: The hybrid neural network model training strategy in step S3 includes: Dataset: Collect battery samples at different aging stages and label the actual SOH values; Loss function: Joint optimization of root mean square error (RMSE) and cross-entropy loss; Optimizer: Adagrad algorithm, with adaptive learning rate adjustment.
8. The method for enhanced diagnosis of energy storage battery health status based on battery body feature analysis according to claim 6, characterized in that: The hybrid neural network model is constructed in step S3 as follows: CNN section: Let the input ultrasound image be... After the ll-th convolutional layer: ; in, Indicates the first The feature map output by the convolution operation is the result after convolution, biasing, and ReLU activation. Indicates the first Input feature maps for layer convolution operations; , The weights and biases of the convolutional kernel in this layer are... This represents the convolution operation. The activation function is used; finally, the image feature vector is obtained through global average pooling. ; BiLSTM section: Processing time-series materials and EIS feature sequences: ； in, Let be the material eigenvector at time t. The eigenvector of the electrochemical impedance spectrum at time t, together with the eigenvector, constitutes the time-series input sequence; Forward LSTM computes hidden state Backward LSTM computation The final output is obtained by splicing the data at each time step. Finally, the output of the last time step is taken or temporal pooling is performed to obtain the temporal feature vector. ; Feature fusion and output: and Concatenate, input fully connected layer: ； in, , These are the parameters for the fully connected layer. It is a linear activation function, and its output is a health status indicator; Hyperparameter optimization: The learning rate, number of network layers, and number of neurons are optimized using the genetic algorithm (GA). The fitness function is defined as the reciprocal of the model's diagnostic error on the validation set. The optimal combination of hyperparameters is searched iteratively through selection, crossover, and mutation operations.
9. The enhanced diagnostic method for the health status of energy storage batteries based on battery body feature analysis according to claim 1, characterized in that: The diagnostic results in step S4 also include quantitative indicators of battery health status and dynamic threshold warnings.
10. The method for enhanced diagnosis of energy storage battery health status based on battery body feature analysis according to claim 9, characterized in that: Step S4 specifically includes: The Comprehensive Health Index (CHI), as the main output of the model, can be defined as: ； in, For sub-indicators of health status, the model directly outputs the values; For the normalized thermal runaway risk score, For consistency scoring functions of other features, , , The weighting coefficients and ; Dynamic threshold warning: Warning threshold T alarm It can be dynamically adjusted according to the working conditions: ； in, Based on the threshold, , For temperature and charge / discharge rate Adjustment coefficient.