Lithium ion battery life loss evaluation method and device, medium and equipment

Abstract

The invention relates to the technical field of lithium ion battery testing, and discloses a lithium ion battery life loss evaluation method and device, a medium and equipment, and the method comprises the following steps: S1, collecting multi-modal dynamic data of a lithium ion battery in a charge-discharge cycle process; s2, performing time-frequency domain conjoint analysis on the multi-modal dynamic data, and extracting a battery aging sensitive feature set; s3, constructing a multi-scale coupling model of battery life loss; by constructing a multi-modal data fusion mechanism and a dynamic feature extraction system, during lithium ion battery life loss evaluation, change trends of electrochemical impedance spectroscopy and heat distribution key parameters are captured in real time based on time-frequency domain conjoint analysis, sensor signal distortion and drift problems can be identified, the extraction precision of aging sensitive features is improved, and the accuracy of lithium ion battery life loss evaluation is improved. The problem of characteristic errors caused by signal interference in traditional evaluation is solved, and the accuracy and reliability of life loss evaluation are ensured.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery testing technology, specifically to a method, apparatus, medium, and equipment for assessing the lifespan loss of lithium-ion batteries. Background Technology

[0002] Lithium-ion batteries are rechargeable batteries that primarily function by the movement of lithium ions between the positive and negative electrodes. They possess advantages such as high energy density, high rated voltage, low energy density, strong adaptability to high and low temperatures, and environmental friendliness. With the continuous development of new energy sources, lithium-ion batteries are widely used in household appliances, mobile phones, computers, electric vehicles, and other fields. Due to their long cycle life, high energy density, and low self-discharge, lithium-ion batteries are widely used in electronic products, automobiles, and energy storage devices.

[0003] Currently, due to the complex multimodal data acquisition involved in the lithium-ion battery life loss assessment process, the sensor array equipped with the battery is subject to electrochemical noise interference when extracting battery aging features. It is impossible to detect in real time whether signal distortion and sensor drift occur at the data acquisition points. When the signal distortion phenomenon is not identified, it will cause a large error in the extraction of aging-sensitive features, and the accuracy of life loss assessment cannot be guaranteed.

[0004] Therefore, a method, apparatus, medium, and equipment for assessing the life loss of lithium-ion batteries are proposed to solve the above problems. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method, apparatus, medium, and equipment for assessing the lifespan loss of lithium-ion batteries, thus solving the problems mentioned in the background section.

[0006] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a method, apparatus, medium, and equipment for assessing the lifespan loss of lithium-ion batteries, wherein the method includes the following steps: S1. Collect multimodal dynamic data of lithium-ion batteries during charge-discharge cycles, including electrochemical impedance spectroscopy data, voltage-current response curves, surface temperature distribution data, and internal micro-short circuit signals. S2. Perform time-frequency domain joint analysis on the multimodal dynamic data to extract the battery aging sensitive feature set, including the anode SEI film growth coefficient, cathode active material phase transition gradient, electrolyte decomposition rate and lithium deposition spatial distribution entropy value. S3. Construct a multi-scale coupled model of battery life loss, and map the aging-sensitive feature set to the battery capacity decay trajectory through a convolutional gated recurrent neural network to generate an initial life loss assessment matrix. S4. Dynamically correct the initial life loss assessment matrix based on the battery's historical operating environment parameters, including environmental temperature fluctuation compensation, charge / discharge rate stress weighting and storage time decay factor calibration, to generate environmental adaptive life loss assessment data. S5. Using a transfer reinforcement learning algorithm, the environmental adaptive life loss assessment data is matched with the standard aging database across operating conditions, and the remaining battery life probability distribution curve and critical failure risk threshold are output. S6. Trigger a battery health status classification early warning signal based on the critical failure risk threshold, and generate a lithium-ion battery life loss assessment report.

[0007] Preferably, S1 includes: S11. Real-time acquisition of dynamic response data during battery charging and discharging through an embedded multi-source sensor array, with a sampling frequency higher than 10kHz; S12. The wavelet packet decomposition algorithm is used to perform multi-resolution noise filtering on the dynamic response data to separate the fundamental component of the electrochemical impedance spectrum, the transient component of ohmic polarization, and the slow-change component of concentration polarization.

[0008] Preferably, S2 includes: S21. Based on the Hilbert-Huang transform, perform intrinsic mode decomposition on the fundamental component of the electrochemical impedance spectroscopy to extract the anode interface capacitance attenuation rate. and the increase in cathode charge transfer resistance ; S22. The voltage-current response curve is converted into a spatiotemporal feature tensor through Gram angle field encoding, and local polarization abrupt change features are extracted using a three-dimensional convolution kernel. .

[0009] Preferably, S3 includes: S31. Design a dual-channel convolutional gated recurrent neural network, with the aging-sensitive feature set as input to the first channel. The second channel inputs battery cycle count and cumulative capacity throughput; S32. Dynamically weight the feature contribution of different aging stages through the attention mechanism, and output the capacity decay trajectory function:

[0010] Preferably, S4 includes: S41. Establish the ambient temperature-capacity decay transfer function; S42. The transfer function is subjected to online parameter identification using the Ziegler-Nichols tuning method to generate a capacity decay trajectory after temperature compensation.

[0011] Preferably, S5 includes: S51. Construct the value function of the transition state action; S52. Employ a dual-depth Q-network to optimize the action selection strategy and maximize the cross-condition matching similarity reward value. .

[0012] Preferably, S6 includes: S61. Define the health status grading early warning rule: When the lower limit of the 90% confidence interval of the remaining life probability distribution curve reaches the critical failure threshold. When a red alert is triggered, and the situation is underway... When the interval is reached, a yellow alert is triggered.

[0013] Preferably, the device includes: The multi-source data acquisition module integrates a high-precision voltage and current probe, a multi-frequency EIS excitation source, an infrared thermal imaging unit, and an acoustic emission sensor to acquire battery dynamic response data in real time. The feature fusion processing module, equipped with an FPGA chip, implements a joint time-frequency domain analysis algorithm and outputs an aging-sensitive feature set. The core engine for lifetime prediction is based on a multi-scale coupled model constructed using a GPU-accelerated convolutional gated recurrent neural network. The environmental adaptive correction module dynamically compensates for temperature and charging / discharging stress disturbances through a Kalman filter. The migration decision output module deploys a migration reinforcement learning algorithm library to generate the remaining lifetime probability distribution and early warning signals.

[0014] Preferably, the medium specifically includes: The program instruction set is divided into a data acquisition and control unit, a feature extraction and calculation unit, a neural network training unit, and an early warning strategy generation unit; The data acquisition and control unit is configured to drive the multi-source sensors to synchronously trigger the sampling clock; The feature extraction calculation unit implements parallel operations of Hilbert-Huang transform and Gram angle field coding; The neural network training unit supports online updating of the convolutional kernel weight matrix. And gate parameters of the gated loop unit ; The early warning strategy generation unit has a built-in Bayesian risk decision tree and outputs a three-level health status code.

[0015] Preferably, the equipment includes: Rack-mount main unit, battery test fixture cluster, liquid cooling temperature control system and evaluation device; The main unit has a built-in industrial-grade server and is equipped with a multi-threaded task scheduler to manage program execution on the storage medium. The battery test fixture integrates a probe matrix, which supports the simultaneous evaluation of the life loss of 12 battery modules. The liquid cooling temperature control system uses PID closed-loop control to maintain the test environment temperature within a fluctuation range of ±0.5℃.

[0016] (III) Beneficial Effects Compared with the prior art, the present invention provides a method, apparatus, medium and equipment for assessing the life loss of lithium-ion batteries, which has the following beneficial effects: 1. In this invention, by constructing a multimodal data fusion mechanism and a dynamic feature extraction system, when assessing the life loss of lithium-ion batteries, the changing trends of electrochemical impedance spectroscopy and key parameters of thermal distribution are captured in real time based on joint analysis in the time and frequency domains. This can identify sensor signal distortion and drift problems, improve the extraction accuracy of aging-sensitive features, solve the feature error problem caused by signal interference in traditional assessments, and ensure the accuracy and reliability of life loss assessment.

[0017] 2. In this invention, by using environmental parameter adaptive correction technology, a dynamic transfer function model is established and the attenuation factor is adjusted online when compensating for temperature fluctuations and charging / discharging stress. The environmental parameters are detected in real time to see if they exceed the correction range, thus solving the problem of model mismatch risk. When abnormal stress disturbances are detected, the compensation algorithm is automatically triggered to calibrate the capacity attenuation trajectory, avoiding the accumulation of prediction bias and ensuring the robustness and continuity of the evaluation model in multiple scenarios.

[0018] 3. In this invention, by using transfer reinforcement learning and a hierarchical early warning mechanism, the database mapping relationship is dynamically optimized based on a cross-operating condition matching strategy when predicting remaining battery life, thus solving the problem of data mismapping caused by differences in battery materials and changing usage scenarios. By combining probability distribution curves and three-level risk thresholds, hierarchical early warning of battery health status is achieved, reducing the false alarm rate and improving the real-time performance and decision reliability of failure early warning. Attached Figure Description

[0019] Figure 1 This is a flowchart of a lithium-ion battery life loss assessment method according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] For specific implementation examples, please refer to: Figure 1 A method, apparatus, medium, and equipment for assessing the lifespan loss of lithium-ion batteries, the method comprising the following steps: S1. Collect multimodal dynamic data of lithium-ion batteries during charge-discharge cycles, including electrochemical impedance spectroscopy data, voltage-current response curves, surface temperature distribution data, and internal micro-short circuit signals. S2. Perform time-frequency domain joint analysis on multimodal dynamic data to extract battery aging sensitive feature set, including anode SEI film growth coefficient, cathode active material phase transition gradient, electrolyte decomposition rate and lithium deposition spatial distribution entropy value. S3. Construct a multi-scale coupled model of battery life loss, and use a convolutional gated recurrent neural network to map the aging-sensitive feature set to the battery capacity decay trajectory to generate an initial life loss assessment matrix. S4. Dynamically correct the initial life loss assessment matrix based on the battery's historical operating environment parameters, including environmental temperature fluctuation compensation, charge and discharge rate stress weighting and storage time decay factor calibration, to generate environmentally adaptive life loss assessment data. S5. Using a transfer reinforcement learning algorithm, the environmental adaptive life loss assessment data is matched with the standard aging database across operating conditions, and the remaining battery life probability distribution curve and critical failure risk threshold are output. S6. Trigger a battery health status classification early warning signal based on the critical failure risk threshold, and generate a lithium-ion battery life loss assessment report.

[0022] S1 includes: S11. Real-time acquisition of dynamic response data during battery charging and discharging through an embedded multi-source sensor array, with a sampling frequency higher than 10kHz; S12. Wavelet packet decomposition algorithm is used to perform multi-resolution noise filtering on the dynamic response data, separating the fundamental component of electrochemical impedance spectroscopy, the transient component of ohmic polarization, and the slow-varying component of concentration polarization. The energy threshold of wavelet packet decomposition is as follows: satisfy: in This is the energy threshold determination value. The number of decomposition layers, This is the current decomposition layer number. For the first Layer wavelet coefficients, It is a second-order differential operator. This represents the noise standard deviation.

[0023] S2 includes: S21. Based on the Hilbert-Huang transform, perform intrinsic mode decomposition on the fundamental component of the electrochemical impedance spectroscopy to extract the anode interface capacitance attenuation rate. and the increase in cathode charge transfer resistance This includes the following steps: S211, for the fundamental components of electrochemical impedance spectroscopy Perform intrinsic mode decomposition (EMD) to generate a set of eigenmode components. , of which The highest frequency, The lowest frequency is met, and the screening criteria are satisfied: in This represents the total sampling time for electrochemical impedance spectroscopy. Let be the upper envelope function of the candidate IMF component. Let be the lower envelope function of the candidate IMF component. For time variables, To filter tolerance thresholds; The specific screening process when the minimum value is less than the screening tolerance threshold includes: I. The fundamental component of the electrochemical impedance spectroscopy Repeat the following operations: Extract candidate components; Calculate its upper / lower envelope function and ; Calculate the envelope symmetry error: in For envelope symmetry error, For time indexing, The total sampling time, The mean of the envelope; II. Conditional Judgment: The current candidate component meets the IMF criteria and is accepted as a valid component. Add to the set; Continue iteratively decomposing the candidate components, including adjusting the envelope and reinitializing, until... ; III. Termination Conditions: When the remaining signal is a monotonic trend term, the decomposition stops and the final IMF set is output. S212. Separate low-frequency components from effective IMF components. and mid-to-high frequency components Low frequency components The energy percentage is greater than 65% of the total energy, characterizing the dynamic changes of the anode SEI film; Mid-to-high frequency components The instantaneous frequency is greater than Hz, characterizing the cathode charge transfer process; S213. Calculate the anode interface capacitance decay rate: in The attenuation coefficient of the anode material. The anode interface capacitance decay rate. For time normalization factor, For time derivative operators, For time variables, This is the maximum number of points. These are low-frequency intrinsic mode components; S214. Calculate the cathode charge transfer resistance increment: in This is the increment of the cathode charge transfer resistance. For Hilbert transform operators, Take the real part, The cathode interface degradation coefficient These are the mid-to-high frequency intrinsic mode components. It is a time variable; S22. The voltage-current response curve is converted into a spatiotemporal feature tensor through Gram angle field encoding, and local polarization abrupt change features are extracted using a three-dimensional convolution kernel. This includes the following operations: I. Spatiotemporal Feature Tensor Generation: Time-series partitioning: Dividing the voltage-current response curve into fixed time windows. A continuous subsequence; Gram corner field encoding: Perform the following operations for each subsequence: Normalization: This process normalizes the voltage value. and current value Scale to the unit interval [0,1]; Polar coordinate mapping: Calculating angular components and timing radius ; in For angular components, For normalized voltage values, For time radius, This is the time-series index within the subsequence. This represents the total length of the subsequence; Constructing the Gram matrix Its elements satisfy: in For Gram matrix elements, For time points voltage phase angle, For time points The voltage phase angle; Tensor Construction: Stack the Gram matrices of all subsequences along the time axis to form a three-dimensional spatiotemporal feature tensor, with the dimension being time. Voltage-related space The current-related space is defined by the following formula: in For a three-dimensional spatiotemporal feature tensor, For the real number space, For tensor dimension; II. Extraction of Local Polarization Abrupt Change Features: 3D convolution kernel design: Spatial dimension Planar: The weights follow a Gaussian distribution, focusing on the local spatial correlation between voltage and current; Time dimension Axis: Weights are distributed differentially. , enhance the response at signal abrupt change points, among which The convolution kernel vector is the time dimension; Convolution operation: convolution kernel In the spatiotemporal feature tensor The calculation is performed by sliding upwards, generating convolution response values ​​point by point; Negative responses are filtered out using the ReLU activation function, while positive polarity mutation components are preserved. Feature output: The response value after aggregation activation is a local polarization mutation feature. Quantify polarization anomalies during battery charging and discharging.

[0024] S3 includes: S31. Design a dual-channel convolutional gated recurrent neural network, with the first channel inputting the aging-sensitive feature set. The second channel inputs battery cycle count and cumulative capacity throughput; S32. Dynamically weight the feature contributions of different aging stages through an attention mechanism, and output the capacity decay trajectory function: in This is the capacity decay amount. The number of loops. For feature vectors, The ambient temperature.

[0025] The weight allocation function for the attention mechanism is: in for The feature weight vector at time step, For the softmax function, For the hidden state vector, For aging-sensitive feature set, For feature dimension, This represents a vector concatenation operation. This is a trainable weight matrix.

[0026] S4 includes: S41. Establish the ambient temperature-capacity decay transfer function: in For complex frequency variables, The coefficient of thermal stress. Let be the relaxation time constant. The inertial time constant, It is a natural constant; S42. Online parameter identification of the transfer function is performed using the Ziegler-Nichols tuning method to generate the capacity decay trajectory after temperature compensation. Specifically, this includes: definition of the vector of parameters to be identified: in The parameter vector to be identified, The coefficient of thermal stress. Let be the relaxation time constant. The inertial time constant; Predicted capacity value generation: for Performing the inverse Laplace transform yields the time-domain response. The specific operation is as follows: in For time-domain response function, For the inverse Laplace transform operator, For transfer functions, It is a complex frequency variable; calculate Predicted capacity value at time: in To predict capacity values, For the initial capacity, It is a time-domain function of ambient temperature. For the current time, It is a function of ambient temperature; Cost function construction: in Optimize the objective function for the parameters. The parameter vector to be identified, For battery charge-discharge cycles, This is the forgetting decay coefficient. for Predicted capacity value at time 10:00 Depend on pass generate; Parameter identification and compensation: Adjusted using the Ziegler-Nichols tuning method ,minimize ; Use the optimized renew Output capacity decay trajectory after temperature compensation: in For the optimal parameter vector, The operator that minimizes the objective function, Let cost function be This is the parameter vector to be optimized.

[0027] S5 includes: S51. Construct the transition state action value function: in The state-action value function is the transition state. For environmentally adaptive lifetime loss assessment data, action For database matching strategies, For the target policy function, For neural network parameter set; S52. Employ a dual-depth Q-network to optimize the action selection strategy and maximize the cross-condition matching similarity reward value. The specific operations include: Objective function definition: Transition state action value function The optimization objective is: in The value function for transition state actions. This is data for environmentally adaptive lifetime loss assessment. For database matching strategies, For the target policy function, For neural network parameter set; Reward Mechanism: The similarity reward value for cross-working-condition matching is calculated as follows: in This is the current operating condition state vector. This serves as reference status data in the historical operating condition database; The higher the similarity, The larger the value, the better the strategy matching accuracy; By maximizing Drive DDQN to select the optimal action A database matching strategy enables the battery life assessment model to adapt in real time without training under varying operating conditions.

[0028] S6 includes: S61. Define the health status classification and early warning rules: Red alert trigger condition: When the lower limit of the 90% confidence interval of the remaining life probability distribution curve is reached. satisfy: This immediately triggers a red alert, initiating battery system safety lock and maintenance alarm. Yellow alert trigger condition: When the remaining lifespan value satisfy: This triggers a yellow alert, initiating preventative maintenance scheduling and parameter optimization. Green and safe status: When When the condition is deemed healthy, a low-risk indicator is output. S62. Calculate the critical failure threshold. : in This is the dynamic critical failure threshold. The mean of the baseline capacity distribution. The standard deviation of the baseline capacity distribution. This represents the current state of battery health (SOH) value. This is the boundary adjustment factor; S63. Dynamically update early warning thresholds: If the number of charge / discharge cycles exceeds 100 per cycle, recalculate. and Adapting to battery aging and drift; when At that time, Scaling to: ; in For the updated boundary adjustment factor, The original boundary adjustment factor. It is a natural constant. This is the material attenuation coefficient.

[0029] The device includes: The multi-source data acquisition module integrates a high-precision voltage and current probe, a multi-frequency EIS excitation source, an infrared thermal imaging unit, and an acoustic emission sensor to acquire battery dynamic response data in real time. The feature fusion processing module, equipped with an FPGA chip, implements a joint time-frequency domain analysis algorithm and outputs an aging-sensitive feature set. The core engine for lifetime prediction is based on a multi-scale coupled model constructed using a GPU-accelerated convolutional gated recurrent neural network. The environmental adaptive correction module dynamically compensates for temperature and charging / discharging stress disturbances through a Kalman filter. The migration decision output module deploys a migration reinforcement learning algorithm library to generate the remaining lifetime probability distribution and early warning signals.

[0030] The medium specifically includes: the program instruction set is divided into a data acquisition and control unit, a feature extraction and calculation unit, a neural network training unit, and an early warning strategy generation unit; The data acquisition and control unit is configured to drive multiple source sensors to synchronously trigger the sampling clock; The feature extraction computation unit implements parallel computation of Hilbert-Huang transform and Gram angle field coding; The neural network training unit supports online updating of the convolutional kernel weight matrix. And gate parameters of the gated loop unit ; The early warning strategy generation unit has a built-in Bayesian risk decision tree and outputs a three-level health status code.

[0031] The equipment includes: Rack-mount main unit, battery test fixture cluster, liquid cooling temperature control system and evaluation device; The main unit has an industrial-grade server built in, configured with a multi-threaded task scheduler to manage program execution on the storage media; The battery test fixture integrates a probe matrix, supporting the simultaneous evaluation of the life loss of 12 battery modules. The liquid cooling temperature control system uses PID closed-loop control to maintain the test environment temperature within a fluctuation range of ±0.5℃.

[0032] The operating steps of this lithium-ion battery life loss assessment method, apparatus, medium, and equipment are as follows: Step 1: High-precision acquisition of multi-source dynamic data The system utilizes an embedded multi-source sensor array, including a high-precision voltage / current probe, a multi-frequency EIS excitation source, and an infrared thermal imaging unit, to capture in real-time the electrochemical impedance spectroscopy, voltage-current response curves, surface temperature distribution, and internal micro-short-circuit signals during battery charge-discharge cycles. Synchronous trigger sampling technology ensures spatiotemporal alignment of multimodal data, with a sampling frequency exceeding 10kHz to cover the entire dynamic response process of the battery. The acquired raw data undergoes multi-resolution filtering using a wavelet packet decomposition algorithm to separate the fundamental electrochemical impedance component, ohmic polarization transient component, and concentration polarization gradual change component, eliminating high-frequency noise interference.

[0033] Step 2: Time-frequency domain co-extraction of aging-sensitive characteristics Hilbert-Huang transform was performed on the fundamental component of the filtered electrochemical impedance spectroscopy (EIS), and interface degradation parameters such as the anode SEI film capacitance decay rate and cathode charge transfer resistance increment were extracted through adaptive intrinsic mode decomposition. Simultaneously, Gram angle field coding was employed to convert the voltage-current curves into spatiotemporal feature tensors, and three-dimensional convolutional kernels were used to capture local polarization abrupt changes. This step integrates the time-domain transient response and frequency-domain impedance evolution to construct a multi-dimensional sensitive feature set for battery aging.

[0034] Step 3: Construction of a multi-scale coupled lifetime prediction model A dual-channel convolutional gated recurrent neural network is designed. The first channel takes an aging-sensitive feature vector as input, while the second channel takes loop counts and cumulative capacity throughput as inputs. An attention mechanism dynamically allocates feature weights for different aging stages to generate a capacity decay trajectory function. During network training, a gradient truncation strategy is employed to prevent long-term dependency bias. The network outputs an initial evaluation matrix for battery capacity decay, achieving cross-scale correlation modeling between microscopic interface degradation and macroscopic performance degradation.

[0035] Step 4: Dynamic Compensation Calibration of Environmental Stress A transfer function model of ambient temperature-capacity decay was established, and the temperature stress coefficient and relaxation time constant were identified online based on the Ziegler-Nichols tuning method. A time forgetting factor was introduced to perform weighted fitting of the measured capacity trajectory, and a Kalman filter was used to compensate for model drift caused by temperature fluctuations, sudden changes in charge / discharge rates, and rest time in real time. This generated environmentally adaptive lifetime loss assessment data, improving the robustness of the model under varying operating conditions.

[0036] Step 5: Cross-Scenario Migration Early Warning Decision Generation A transfer learning reinforcement learning framework is constructed, using environmental adaptive assessment data as state input. A cross-condition matching strategy for a standard aging database is optimized based on a dual-deep Q-network. The remaining lifetime probability distribution curve is generated by maximizing the transfer similarity reward value, and a critical failure risk threshold is calculated using a Bayesian risk decision tree. A tiered early warning signal is triggered when the probability distribution reaches the 90% confidence interval of the threshold, and an assessment report including health status rating, remaining cycle count, and failure probability is output.

[0037] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0038] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the lifespan loss of a lithium-ion battery, characterized in that: The method includes the following steps: S1. Collect multimodal dynamic data of lithium-ion batteries during charge-discharge cycles, including electrochemical impedance spectroscopy data, voltage-current response curves, surface temperature distribution data, and internal micro-short circuit signals. S2. Perform time-frequency domain joint analysis on the multimodal dynamic data to extract the battery aging sensitive feature set, including the anode SEI film growth coefficient, cathode active material phase transition gradient, electrolyte decomposition rate and lithium deposition spatial distribution entropy value. S3. Construct a multi-scale coupled model of battery life loss, and map the aging-sensitive feature set to the battery capacity decay trajectory through a convolutional gated recurrent neural network to generate an initial life loss assessment matrix. S4. Dynamically correct the initial life loss assessment matrix based on the battery's historical operating environment parameters, including environmental temperature fluctuation compensation, charge / discharge rate stress weighting and storage time decay factor calibration, to generate environmental adaptive life loss assessment data. S5. Using a transfer reinforcement learning algorithm, the environmental adaptive life loss assessment data is matched with the standard aging database across operating conditions, and the remaining battery life probability distribution curve and critical failure risk threshold are output. S6. Trigger a battery health status classification early warning signal based on the critical failure risk threshold, and generate a lithium-ion battery life loss assessment report.

2. The method for assessing the lifespan loss of a lithium-ion battery according to claim 1, characterized in that: S1 includes: S11. Real-time acquisition of dynamic response data during battery charging and discharging through an embedded multi-source sensor array, with a sampling frequency higher than 10kHz; S12. The wavelet packet decomposition algorithm is used to perform multi-resolution noise filtering on the dynamic response data to separate the fundamental component of the electrochemical impedance spectrum, the transient component of ohmic polarization, and the slow-change component of concentration polarization.

3. The method for assessing the lifespan loss of a lithium-ion battery according to claim 1, characterized in that: S2 includes: S21. Based on the Hilbert-Huang transform, perform intrinsic mode decomposition on the fundamental component of the electrochemical impedance spectroscopy to extract the anode interface capacitance attenuation rate. and the increase in cathode charge transfer resistance ; S22. The voltage-current response curve is converted into a spatiotemporal feature tensor through Gram angle field encoding, and local polarization abrupt change features are extracted using a three-dimensional convolution kernel. .

4. The method for assessing the lifespan loss of a lithium-ion battery according to claim 1, characterized in that: S3 includes: S31. Design a dual-channel convolutional gated recurrent neural network, with the aging-sensitive feature set as input to the first channel. The second channel inputs battery cycle count and cumulative capacity throughput; S32. Dynamically weight the feature contribution of different aging stages through an attention mechanism, and output the capacity decay trajectory function.

5. The method for assessing the lifespan loss of a lithium-ion battery according to claim 1, characterized in that: S4 includes: S41. Establish the ambient temperature-capacity decay transfer function; S42. The transfer function is subjected to online parameter identification using the Ziegler-Nichols tuning method to generate a capacity decay trajectory after temperature compensation.

6. The method for assessing the lifespan loss of a lithium-ion battery according to claim 1, characterized in that: S5 includes: S51. Construct the value function of the transition state action; S52. Employ a dual-depth Q-network to optimize the action selection strategy and maximize the cross-condition matching similarity reward value. .

7. The method for assessing the lifespan loss of a lithium-ion battery according to claim 1, characterized in that: S6 includes: S61. Define the health status grading early warning rule: When the lower limit of the 90% confidence interval of the remaining life probability distribution curve reaches the critical failure threshold. When a red alert is triggered, and the situation is underway... When the interval is reached, a yellow alert is triggered.

8. A lithium-ion battery life loss assessment device, characterized in that: A method for assessing the lifespan loss of a lithium-ion battery as described in any one of claims 1-7, characterized in that the apparatus comprises: The multi-source data acquisition module integrates a high-precision voltage and current probe, a multi-frequency EIS excitation source, an infrared thermal imaging unit, and an acoustic emission sensor to acquire battery dynamic response data in real time. The feature fusion processing module, equipped with an FPGA chip, implements a joint time-frequency domain analysis algorithm and outputs an aging-sensitive feature set. The core engine for lifetime prediction is based on a multi-scale coupled model constructed using a GPU-accelerated convolutional gated recurrent neural network. The environmental adaptive correction module dynamically compensates for temperature and charging / discharging stress disturbances through a Kalman filter. The migration decision output module deploys a migration reinforcement learning algorithm library to generate the remaining lifetime probability distribution and early warning signals.

9. A lithium-ion battery life loss assessment medium, wherein a computer program is stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the lithium-ion battery life loss assessment method according to any one of claims 1-7, wherein the medium specifically includes: The program instruction set is divided into a data acquisition and control unit, a feature extraction and calculation unit, a neural network training unit, and an early warning strategy generation unit; The data acquisition and control unit is configured to drive the multi-source sensors to synchronously trigger the sampling clock; The feature extraction calculation unit implements parallel operations of Hilbert-Huang transform and Gram angle field coding; The neural network training unit supports online updating of the convolutional kernel weight matrix. And gate parameters of the gated loop unit ; The early warning strategy generation unit has a built-in Bayesian risk decision tree and outputs a three-level health status code.

10. A lithium-ion battery life loss assessment device, characterized in that the device... include: Rack-mount main unit, battery test fixture cluster, liquid cooling temperature control system and evaluation device; The main unit has a built-in industrial-grade server and is equipped with a multi-threaded task scheduler to manage program execution on the storage medium. The battery test fixture integrates a probe matrix, which supports the simultaneous evaluation of the life loss of 12 battery modules. The liquid cooling temperature control system uses PID closed-loop control to maintain the test environment temperature within a fluctuation range of ±0.5℃.

Images