CN-121978567-A - Energy storage battery health state enhanced diagnosis method based on battery body feature analysis

CN121978567ACN 121978567 ACN121978567 ACN 121978567ACN-121978567-A

Abstract

The invention belongs to the technical field of energy storage battery health state diagnosis, and particularly discloses an energy storage battery health state enhanced diagnosis method based on battery body feature analysis. The method builds a multi-mode mathematical model based on the characteristic analysis of the battery body, breaks through the dependence of the traditional method on external parameters, can realize the enhanced diagnosis of the health state of the energy storage battery, can remarkably improve the diagnosis precision, realize early fault early warning and full life cycle management, and remarkably improve the compatibility and expansibility.

Inventors

HUO XUANMIN
WANG JUANJUAN
WANG JIAHAN
FANG YUAN
LIU YUAN
ZHAO BIAO

Assignees

国网河南省电力公司新乡供电公司

Dates

Publication Date: 20260505
Application Date: 20251219

Claims (10)

1. The energy storage battery health state enhanced diagnosis method based on the battery body feature analysis is characterized by comprising the following steps of: S1, acquiring characteristic data of a battery body, wherein the characteristic data comprise ultrasonic imaging data, material composition data and electrochemical impedance data; S2, preprocessing the characteristic data and extracting the characteristics: s3, constructing a hybrid neural network model based on a two-way long-short-term memory network and a convolutional neural network, and inputting the characteristic data for training; s4, outputting a battery health state diagnosis result by using the trained model, wherein the battery health state diagnosis result comprises a comprehensive health index and each sub-index.
2. The method for strengthening diagnosis of the health state of the energy storage battery based on the characteristic analysis of the battery body according to claim 1, wherein the ultrasonic imaging data in the step S1 are obtained by utilizing a high-frequency ultrasonic transmission technology and comprise sub-millimeter imaging data of electrolyte infiltration state and gas production distribution inside the battery; The material composition data comprises electrode material lattice parameter, particle size and element content, which are obtained by means including X-ray diffraction and scanning electron microscope; Electrochemical impedance data, including charge transfer resistance, ion diffusion coefficient, are resolved by fractional equivalent circuit models.
3. The method for enhanced diagnosis of the health status of the energy storage battery based on the battery body feature analysis according to claim 2, wherein the ultrasonic detection during the acquisition of the ultrasonic imaging data in step S1: Scanning the battery by adopting a high-frequency ultrasonic probe, generating a gray image of an internal structure through a signal processing algorithm, and extracting the number of bubbles and the wetting area characteristics of electrolyte; The ultrasonic imaging data is scanned by a high-frequency ultrasonic probe to receive an echo signal U (t), and after signal processing, an internal structure gray level image I (x, y) is reconstructed by a back projection algorithm or a delay superposition algorithm, wherein (x, y) is an image pixel coordinate; material analysis at material composition data acquisition: periodically collecting battery electrode samples, analyzing crystal structure changes through XRD, observing particle morphology evolution through SEM, and establishing a material degradation database; Obtaining a diffraction pattern of material composition data through X-ray diffraction XRD, and calculating a lattice spacing d by using a Bragg equation, wherein n lambda= dsin θ, lambda is the wavelength of X-ray, θ is the diffraction angle, and n is the diffraction order; EIS test at electrochemical impedance data acquisition: EIS tests are carried out under different SOC states, and FECM models are used for fitting to obtain charge transfer resistance and Warburg impedance parameters; the electrochemical impedance data are subjected to EIS test through an electrochemical workstation to obtain complex impedance spectrum Z (omega) =Z '(omega) + jZ' (omega); Wherein ω is angular frequency, Z' (ω) is real, Z "(ω) is imaginary, j is imaginary unit, i.e., j 2 = -1, fitting using a fractional equivalent circuit model FECM, model expression: ; Wherein, the Is an equivalent series resistance of the resistor, In the form of a charge transfer resistor, In the case of a constant phase angle element, For the fractional order index thereof, Is the Warburg coefficient.
4. The method for strengthening diagnosis of the health state of the energy storage battery based on the characteristic analysis of the battery body according to claim 1, wherein the characteristic preprocessing in the step S2 is to perform standardization and noise reduction on the ultrasonic image, the material composition data and the EIS parameters; the characteristic engineering comprises the following steps: Ultrasonic characteristics including bubble density, infiltration rate and sound velocity attenuation coefficient; the material is characterized by lattice constant, particle size distribution and element content; EIS characteristics are R ct , zw and equivalent series resistance.
5. The method for enhanced diagnosis of the state of health of an energy storage battery based on a battery body profile analysis as set forth in claim 4, wherein the step S2 comprises: S2-1, data standardization, namely, Z-score standardization is carried out on data from different sources and different dimensions; s2-2, ultrasonic feature extraction, which comprises the following steps: Bubble density : ; Wherein, the For the total number of bubble contours identified by the image segmentation algorithm, An area that is a region of interest, ROI; Infiltration rate : ; Wherein, the The area of the electrolyte infiltration area is distinguished by a gray threshold; is the total area of the electrode area; s2-3 material characterization, obtaining lattice constants (a, c) from XRD patterns by peak position fitting, and obtaining average particle size from SEM images by image analysis Standard deviation of distribution thereof ; And S2-4.EIS feature extraction, namely directly acquiring parameter values from the fitted FECM model.
6. The method for strengthening diagnosis of the health state of the energy storage battery based on the battery body feature analysis according to claim 1, wherein the hybrid neural network model in the step S3 optimizes the super parameters by adopting a genetic algorithm, and the hybrid neural network model comprises: The CNN layer is used for carrying out convolution operation on the ultrasonic image and extracting spatial characteristics including defect positions and structural anomalies; BiLSTM layers for processing time sequence characteristics including material degradation trend and EIS parameter change to capture long-term dependence; The full connection layer is used for fusing multi-mode characteristics, outputting CHI and various sub-indexes including SOH and thermal runaway risk.
7. The method for enhanced diagnosis of the state of health of an energy storage battery based on a battery body profile analysis of claim 6, wherein the training strategy of the hybrid neural network model in step S3 comprises: Collecting battery samples at different aging stages, and marking a real SOH value; loss function, namely Root Mean Square Error (RMSE) and cross entropy loss are combined and optimized; And (3) optimizing Adagrad algorithm, and adaptively adjusting the learning rate.
8. The method for enhanced diagnosis of the health status of the energy storage battery based on the battery body feature analysis of claim 6, wherein the hybrid neural network model in step S3 is constructed as follows: CNN part, let the input ultrasound image be Through the ll layer convolution operation: ; Wherein, the Represent the first The feature map output by the layer convolution operation, i.e. the result after convolution, biasing and ReLU activation functions, Represent the first An input feature map of a layer convolution operation; 、 The kernel weights and offsets are convolved for this layer, A convolution operation is represented and is performed, Finally obtaining the image characteristic vector by global average pooling ; Part BiLSTM processing time series materials and EIS characteristic sequences: ; Wherein, the Is the characteristic vector of the material at the time t, Is the characteristic vector of the electrochemical impedance spectrum at the time t, the two together form a time sequence input sequence; Forward LSTM computation hidden state Backward LSTM computation Splicing at each time step to obtain final output Finally, the output of the last time step is taken or time sequence pooling is carried out to obtain time sequence characteristic vector ; Feature fusion and output And (3) with Splicing, and inputting a full connection layer: ; Wherein, the 、 Is a parameter of the full connection layer, The output is a health state index as a linear activation function; And the super-parameter optimization comprises the steps of optimizing the super-parameters of the learning rate, the network layer number and the neuron number by using a genetic algorithm GA, defining a fitness function as the inverse of the diagnosis error of the model on the verification set, and iteratively searching the optimal super-parameter combination through selection, crossing and mutation operations.
9. The method for enhanced diagnosis of the state of health of an energy storage battery based on the analysis of the body characteristics of the battery according to claim 1, wherein the diagnosis result in step S4 further comprises a quantitative indicator of the state of health of the battery and a dynamic threshold early warning.
10. The method for enhanced diagnosis of the state of health of an energy storage battery based on a battery body profile analysis as set forth in claim 9, wherein the step S4 specifically includes: The comprehensive health index CHI is taken as a model main output and can be defined as: ; Wherein, the The model directly outputs a value for the sub-index of the health state; for the normalized thermal runaway risk score, For the consistency scoring function of other features, , , Is a weighting coefficient and ; Dynamic threshold early warning, wherein the early warning threshold T alarm can be dynamically adjusted according to working conditions: ; Wherein, the As a basis for the threshold value, 、 Is the temperature Charge-discharge multiplying power Is used for the adjustment of the coefficient of (c).

Description

Energy storage battery health state enhanced diagnosis method based on battery body feature analysis Technical Field The invention belongs to the technical field of energy storage battery health state diagnosis, and particularly relates to an energy storage battery health state enhanced diagnosis method based on battery body feature analysis. Background Current state of health (SOH) diagnostics of energy storage cells rely primarily on external parameters (e.g., voltage, current, temperature) and traditional electrochemical methods (e.g., electrochemical impedance spectroscopy EIS). However, these methods have significant drawbacks: 1. The external parameters are limited in that signals such as voltage and current are easy to be interfered by working conditions, and the body characteristics such as degradation of materials in the battery, drying of electrolyte and the like cannot be directly reflected. 2. The traditional electrochemical method has the defects that the EIS needs to be tested offline and cannot be monitored in real time, and the disassembly analysis is destructive detection and cannot be used for a battery in operation. 3. The model precision is insufficient, the existing data driving model (such as a support vector machine) only depends on external parameters, and the complex physicochemical process inside the battery is difficult to capture. At present, although ultrasonic scanning imaging technology developed by the university of Huazhong science and technology team can detect states of gas production, electrolyte infiltration and the like in a battery in real time, the ultrasonic scanning imaging technology is not deeply fused with other body characteristics (such as material components and microstructures) and mathematical models, so that the diagnosis dimension is single. Disclosure of Invention In order to solve the defects in the prior art, the invention aims to provide the energy storage battery health status reinforced diagnosis method based on the battery body feature analysis, which is used for constructing a multi-mode mathematical model based on the battery body feature analysis, breaking through the dependence of the traditional method on external parameters and realizing the reinforced diagnosis of the energy storage battery health status. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: An energy storage battery health status enhanced diagnosis method based on battery body feature analysis comprises the following steps: S1, acquiring characteristic data of a battery body, wherein the characteristic data comprise ultrasonic imaging data, material composition data and electrochemical impedance data; S2, preprocessing the characteristic data and extracting the characteristics: s3, constructing a hybrid neural network model based on a two-way long-short-term memory network and a convolutional neural network, and inputting the characteristic data for training; s4, outputting a battery health state diagnosis result by using the trained model, wherein the battery health state diagnosis result comprises a comprehensive health index and each sub-index. Further, the ultrasonic imaging data in the step S1 are obtained by utilizing a high-frequency ultrasonic transmission technology, and the ultrasonic imaging data comprise sub-millimeter imaging data of electrolyte infiltration state and gas production distribution inside a battery; The material composition data comprises electrode material lattice parameter, particle size and element content, which are obtained by means including X-ray diffraction and scanning electron microscope; Electrochemical impedance data, including charge transfer resistance, ion diffusion coefficient, are resolved by fractional equivalent circuit models. Further, in the step S1, ultrasonic detection is performed during ultrasonic imaging data acquisition, wherein a high-frequency ultrasonic probe is adopted to scan a battery, a gray level image of an internal structure is generated through a signal processing algorithm, and the characteristics of the number of bubbles and the wetting area of electrolyte are extracted; The ultrasonic imaging data is scanned by a high-frequency ultrasonic probe to receive an echo signal U (t), and after signal processing, an internal structure gray level image I (x, y) is reconstructed by a back projection algorithm or a delay superposition algorithm, wherein (x, y) is an image pixel coordinate; The material analysis during the material composition data acquisition comprises periodically collecting battery electrode samples, analyzing crystal structure change by XRD, observing particle morphology evolution by SEM, and establishing a material degradation database; Obtaining a diffraction pattern of material composition data through X-ray diffraction XRD, and calculating a lattice spacing d by using a Bragg equation, wherein n lambda= dsin θ, lambda is the wavelength of X-ray, θ is