A method for quickly detecting a power lithium battery health critical retirement period
By collecting multi-dimensional data at high frequency and building a machine learning model, combined with a dual-threshold judgment mechanism and confidence interval output, the problem of rapid and accurate detection of the critical retirement period of power lithium batteries is solved, improving detection accuracy and safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINESE PEOPLES LIBERATION ARMY UNIT 68302
- Filing Date
- 2026-03-26
- Publication Date
- 2026-06-23
AI Technical Summary
Existing methods for detecting the health status of power lithium batteries lack accuracy and adaptability during the critical retirement period, making it difficult to quickly and accurately identify battery performance degradation, leading to safety hazards or waste of resources.
High-frequency acquisition of multi-dimensional state data, including terminal voltage, charging and discharging current, battery surface temperature and internal impedance spectrum, is employed. Combined with wavelet denoising and outlier removal techniques, a multi-dimensional feature fusion model based on machine learning is constructed. A transfer learning mechanism is introduced, and accurate detection is achieved through a dual threshold judgment mechanism and confidence interval output.
It enables rapid and accurate detection of the critical retirement period of power lithium batteries, reduces misjudgments, improves the scientific nature of BMS decision-making, and balances battery safety and economy.
Smart Images
Figure CN122260128A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of lithium battery health testing technology, specifically to a rapid detection method for the critical retirement period of power lithium batteries. Background Technology
[0002] With the rapid development of the new energy vehicle industry and the widespread application of energy storage systems, the health management and retirement assessment of power lithium batteries, as core energy carriers, throughout their entire life cycle has become a focus of industry attention. During use, power lithium batteries gradually experience capacity decay and increased internal resistance. When their State of Health (SOH) drops to a certain level (typically 80%), they enter the "critical retirement period." At this stage, battery performance degradation exhibits significant non-linear characteristics. If it cannot be identified and retired in a timely and accurate manner, it can easily lead to safety hazards such as thermal runaway; conversely, premature retirement will result in huge resource waste and economic losses. Therefore, how to achieve rapid and accurate detection of the critical retirement period of power lithium batteries is a key technical problem that urgently needs to be solved in the current field of battery management systems (BMS).
[0003] However, existing battery SOH detection and decommissioning assessment technologies still have the following significant shortcomings, making it difficult to meet practical engineering needs: 1. Traditional detection methods lack accuracy and adaptability during the critical period: Currently, the mainstream SOH estimation methods mainly include the ampere-hour integration method, the open-circuit voltage (OCV) method, and filtering algorithms based on equivalent circuit models (ECM) (such as EKF and UKF). The ampere-hour integration method is affected by the accuracy of current sampling and cumulative error, resulting in large long-term operating errors; the OCV method requires a long period of rest to reach electrochemical equilibrium, which cannot meet the needs of "rapid detection"; although ECM-based algorithms can run online, in the later stages of battery aging (critical retirement period), the internal electrochemical characteristics of the battery undergo drastic changes (such as lithium deposition and SEI film thickening), and the model parameters (such as polarization capacitance and Warburg impedance) are severely mismatched with the actual physical characteristics, resulting in a significant increase in SOH estimation error (usually >5%), making it impossible to accurately capture the small changes at the critical point.
[0004] 2. Limited Feature Extraction Dimensions and Lack of In-Depth Exploration of Microscopic Degradation Mechanisms: Existing data-driven detection methods mostly utilize only macroscopic physical quantities such as terminal voltage, current, and surface temperature as input features. However, during the critical period of battery health, changes in macroscopic parameters often lag behind the degradation of internal microstructures. For example, the shift of characteristic peaks and changes in the voltage relaxation time constant in differential voltage curves (DVA) and differential capacity curves (ICA) are sensitive indicators reflecting lithium-ion diffusion capacity and active lithium loss, but existing technologies rarely incorporate these high-dimensional sensitive features (such as the dV / dQ peak slope, internal resistance growth rate ηR, and relaxation time constant τ) into real-time detection models. This results in a sluggish perception of the "critical state," often issuing an alarm only after the battery has severely aged, thus losing its warning significance. Summary of the Invention
[0005] To address the aforementioned technical problems, a rapid detection method for the critical retirement period of power lithium batteries is provided. This technical solution solves the problems mentioned in the background section.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect of the present invention, a method for rapid detection of the critical retirement period of a power lithium battery is provided, comprising: During the charging and discharging process of the battery, multi-dimensional state data of the battery are collected in real time at a preset high-frequency sampling period. The multi-dimensional state data includes terminal voltage, charging and discharging current, battery surface temperature, internal impedance spectrum and single cell voltage difference. The collected raw data is cleaned and preprocessed, abnormal jump points are removed and timestamps are aligned to generate a standard time series dataset; Based on the standard time series dataset, sensitive characteristic parameters of the battery under critical health conditions are extracted. These sensitive characteristic parameters include the peak slope of the differential voltage curve, the peak offset of the differential capacity curve, the ohmic internal resistance growth rate, and the voltage relaxation characteristic time constant. A fast battery health estimation model based on multi-dimensional feature fusion is constructed. The extracted sensitive feature parameters are input into the model, and the current battery health assessment value and confidence interval are output. Based on the health assessment value and its rate of change, combined with the preset retirement threshold window, it is determined whether the battery has entered the critical retirement period and an alarm signal or retirement suggestion is output.
[0007] Preferably, the real-time acquisition of multi-dimensional state data of the battery at a preset high-frequency sampling period specifically includes: Data acquisition is performed when the battery is in dynamic operating conditions, such as DST or FUDS conditions, or under short-term high-rate charge and discharge pulse excitation. A sliding window mechanism is adopted, with a window length of T and 10s≤T≤60s. Within the window, transient response data of voltage and current are collected at a frequency of milliseconds. Simultaneously, thermal imaging data of the battery module is collected, and the temperature field distribution gradient on the battery surface is obtained through infrared thermometry, serving as an auxiliary data source for thermal runaway early warning.
[0008] Preferably, the cleaning and preprocessing of the collected raw data specifically includes: The voltage and current signals are denoised using a wavelet threshold denoising algorithm with 3-5 decomposition layers, and the db4 or sym5 wavelet basis is selected. Use Hampel identifiers or the 3σ criterion to identify and eliminate outliers caused by poor sensor contact or electromagnetic interference. For missing data, K-nearest neighbor-based interpolation or linear interpolation is used to fill in the missing data to ensure the continuity and equal spacing of the data sequence.
[0009] Preferably, the step of extracting sensitive feature parameters of the battery under critical health conditions based on the standard time-series dataset specifically includes: Polynomial fitting was performed on the standard charge-discharge voltage curve to calculate the first derivative of voltage with respect to capacity. and the first derivative of capacitance with respect to voltage Extract the amplitude, area, and half-width of the characteristic peak, and calculate the offset of the characteristic peak relative to the initial state. and ,in, , For the first Voltage values at each sampling point For the first The cumulative capacity value of each sampling point For capacity intervals; Based on pulse charge-discharge data, the ohmic internal resistance is identified online using the equivalent circuit model ECM combined with the recursive least squares method. and polarization resistance Calculate the relative growth rate of internal resistance The formula is: ; in, For the ohmic internal resistance currently identified online, This refers to the initial ohmic internal resistance specified by the battery at the factory. During the resting period after charging and discharging is cut off, voltage recovery curves are collected and fitted with exponential decay functions. Extracting the time constant As a characterization parameter for lithium-ion diffusion ability, among which, This is the steady-state voltage at the end of the resting period. For voltage recovery amplitude, This refers to the settling time.
[0010] Preferably, the construction of a fast battery health estimation model based on multi-dimensional feature fusion specifically includes: Establish a regression model based on machine learning, and select support vector machine, random forest or long short-term memory network as the base learner; The extracted sensitive feature parameters are used as the input vector. Defined as: ; in, This is the peak value of the differential voltage. For differential capacity peak value, The internal resistance growth rate, The voltage relaxation time constant is The standard deviation of the temperature field; The actual SOH value calibrated in the laboratory will be used as the label. ; The hyperparameters of the model are optimized using grid search or Bayesian optimization algorithm, and cross-validation is used to prevent overfitting. By introducing a transfer learning mechanism, the model is pre-trained using historical full life cycle data of batteries of the same model and batch, and then fine-tuned using a small amount of measured data of the current battery, achieving high-precision estimation with small samples.
[0011] Preferably, the output of the current battery health assessment value and confidence interval specifically includes: An ensemble learning framework is used to run multiple different sub-models simultaneously, such as SVM, GPR, and XGBoost. Based on the historical performance of each sub-model on the validation set, dynamically allocate weight coefficients. ,satisfy The final result is obtained by weighted summation. The point estimate is calculated using the following formula: ; in, The number of sub-models, Let i be the predicted value of the i-th sub-model. For the corresponding weights; Calculate the variance or standard deviation of the prediction results of each sub-model, and combine it with the prediction variance of the Gaussian process regression to construct a confidence interval with a 95% confidence level. Specifically: ; in, The standard deviation of the predicted values; If the confidence interval width Greater than the preset threshold If the current data quality is insufficient, a re-collection command will be triggered.
[0012] Preferably, the logic for determining whether a battery has entered a critical health retirement period is as follows: Define the critical retirement period for health as It is in the 80%-85% range and the rate of decline is accelerating; A dual threshold determination mechanism is set up, the First threshold (hard threshold): when If ≤80%, it will be directly judged as retirement; Second threshold (soft threshold / early warning threshold): When 80% < When ≤85%, calculate the most recent A cycle attenuation slope The calculation formula is: ; If the attenuation slope Greater than the historical average decay rate of times, and ≥1.5, that is, it satisfies > And the internal resistance growth rate If it exceeds 30%, it is determined to enter the "critical warning state"; Introduce a hysteresis comparison mechanism to prevent Frequent alarms caused by fluctuations around the threshold, hysteresis width Set it to 1%-2%.
[0013] Preferably, it also includes a trend prediction step based on historical trajectories: Retrieve historical charge-discharge cycle data stored in the battery management system and establish The degradation trajectory varies with the number of cycles or mileage; Using a sliding time window, such as the last 50 cycles, linear or nonlinear fitting is performed on the degradation trajectory to calculate the current degradation acceleration; Combining the Arrhenius equation, based on the current average operating temperature And depth of charge / discharge, predicting the short-term future Evolution trend, decay rate constant Expressed as: ; in, Pre-exponential factor, For activation energy, For molar gas constant, This refers to the average operating temperature. If the prediction If the value is about to fall below the retirement threshold within a short period, an "imminent retirement" warning will be triggered ahead of time.
[0014] Preferably, it also includes cloud-based big data-assisted correction steps: The current battery's feature parameter vector is uploaded to the cloud server and compared with the full life cycle data of batteries of the same model and production batch in the cloud database; Using dynamic time warping The algorithm calculates the similarity between the current battery degradation curve and the standard retirement curve. The calculation formula is: ; in, This is the current battery feature sequence. For standard battery characteristic sequences, To standardize the path, The distance is Euclidean. If the similarity is higher than the preset value, such as 0.9, the remaining lifetime corresponding to the standard curve is directly used as the reference value to correct the prediction results of the local model and realize multi-source data fusion diagnosis.
[0015] In a second aspect of the invention, a rapid detection system for the critical health retirement period of a power lithium battery is also provided, comprising: The acquisition module is used to acquire multi-dimensional state data of the battery in real time during the charging and discharging process at a preset high-frequency sampling period. The multi-dimensional state data includes terminal voltage, charging and discharging current, battery surface temperature, internal impedance spectrum and single cell voltage difference. The preprocessing module is used to clean and preprocess the collected raw data, remove abnormal jump points and align the timestamps to generate a standard time series dataset. The extraction module is used to extract sensitive characteristic parameters of the battery under critical health conditions based on the standard time series dataset. The sensitive characteristic parameters include the peak slope of the differential voltage curve, the peak offset of the differential capacity curve, the ohmic internal resistance growth rate, and the voltage relaxation characteristic time constant. The output module is used to construct a fast battery health estimation model based on multi-dimensional feature fusion, input the extracted sensitive feature parameters into the model, and output the current battery health assessment value and confidence interval. The determination module is used to determine whether the battery has entered the critical retirement period based on the health assessment value and its rate of change, combined with a preset retirement threshold window, and output an alarm signal or retirement suggestion.
[0016] Compared with the prior art, the present invention provides a rapid detection method for the critical retirement period of power lithium batteries, which has the following beneficial effects: This invention abandons the single macroscopic parameter approach, employing high-frequency acquisition of multi-dimensional data such as voltage, current, temperature, and impedance spectrum. Combined with wavelet denoising and outlier removal techniques, it effectively filters electromagnetic interference and sensor noise, generating high-quality standard time-series data to ensure the reliability of subsequent analyses. It introduces sensitive features such as differential voltage / capacity (DVA / ICA) analysis, ohmic resistance growth rate, and voltage relaxation time constant to accurately capture microscopic electrochemical changes such as decreased lithium-ion diffusion capacity and loss of active lithium, overcoming the shortcomings of traditional ampere-hour integration methods with large cumulative errors during the critical period. Furthermore, it constructs a machine learning-based approach (such as LS...) A multi-dimensional feature fusion model (TM, SVM) is used, and a transfer learning mechanism is introduced to achieve high-precision SOH estimation with small samples. By outputting confidence intervals, the uncertainty of prediction is quantified to avoid misjudgment caused by data fluctuations and improve the scientific nature of BMS decision-making. A dual threshold judgment mechanism of "hard threshold + soft threshold" is adopted, which combines the SOH decay slope and the internal resistance growth rate to accurately identify the degradation acceleration inflection point in the 80%-85% range. The Arrhenius equation is used to predict the future short-term degradation trend, realizing the transformation from "passive alarm" to "active early warning" and effectively balancing battery safety and economy. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the method flow for S101-S105 in this invention; Figure 2 This is a schematic diagram of the method flow for S201-S203 in this invention; Figure 3 This is a schematic diagram of the method flow of S301-S303 in this invention; Figure 4 This is a schematic diagram of the method flow for S401-S404 in this invention; Figure 5 This is a schematic diagram of the method flow for S501-S504 in this invention. Detailed Implementation
[0018] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0019] Example 1 Please refer to Figure 1 As shown, in a first aspect of the present invention, a method for rapid detection of the critical retirement period of a power lithium battery is provided, comprising: S101. During the charging and discharging process of the battery, multi-dimensional state data of the battery are collected in real time at a preset high-frequency sampling period. The multi-dimensional state data includes terminal voltage, charging and discharging current, battery surface temperature, internal impedance spectrum and single cell voltage difference. S102. Clean and preprocess the collected raw data, remove abnormal jump points and align the timestamps to generate a standard time series dataset. S103. Based on the standard time series dataset, extract the sensitive characteristic parameters of the battery under critical health conditions. The sensitive characteristic parameters include the peak slope of the differential voltage curve, the peak offset of the differential capacity curve, the growth rate of the ohmic internal resistance, and the voltage relaxation characteristic time constant. S104. Construct a fast battery health estimation model based on multi-dimensional feature fusion, input the extracted sensitive feature parameters into the model, and output the current battery health assessment value and confidence interval. S105. Based on the health assessment value and its rate of change, combined with the preset retirement threshold window, determine whether the battery has entered the critical retirement period and output an alarm signal or retirement suggestion.
[0020] As will be understood by those skilled in the art, this invention abandons the use of single macroscopic parameters and employs high-frequency acquisition of multi-dimensional data such as voltage, current, temperature, and impedance spectra. Combined with wavelet denoising and outlier removal techniques, it effectively filters electromagnetic interference and sensor noise, generating high-quality standard time-series data to ensure the reliability of subsequent analyses. It introduces sensitive features such as differential voltage / capacity (DVA / ICA) analysis, ohmic resistance growth rate, and voltage relaxation time constant to accurately capture microscopic electrochemical changes such as decreased lithium-ion diffusion capacity and loss of active lithium, overcoming the shortcomings of traditional ampere-hour integration methods with large cumulative errors during the critical period. Furthermore, it constructs a machine-based... The system employs a multi-dimensional feature fusion model based on machine learning (such as LSTM and SVM) and introduces a transfer learning mechanism to achieve high-precision SOH estimation with small sample sizes. By outputting confidence intervals, it quantifies prediction uncertainty, avoids misjudgments caused by data fluctuations, and improves the scientific nature of BMS decision-making. It adopts a dual threshold determination mechanism of "hard threshold + soft threshold" and combines the SOH decay slope with the internal resistance growth rate to accurately identify the degradation acceleration inflection point in the 80%-85% range. It uses the Arrhenius equation to predict future short-term degradation trends, realizing the transformation from "passive alarm" to "active early warning" and effectively balancing battery safety and economy.
[0021] Please refer to Figure 2 As shown, multi-dimensional state data of the battery is collected in real time at a preset high-frequency sampling period, specifically including: S201. Data acquisition is performed when the battery is in dynamic operating conditions, such as DST or FUDS conditions, or under short-term high-rate charge and discharge pulse excitation. S202. A sliding window mechanism is adopted, with a window length of T and 10s≤T≤60s. Within the window, transient response data of voltage and current are collected at a frequency of milliseconds. S203. Simultaneously, thermal imaging data of the battery module is collected, and the temperature field distribution gradient on the battery surface is obtained through infrared thermometry, serving as an auxiliary data source for thermal runaway early warning.
[0022] Please refer to Figure 3 As shown, the collected raw data undergoes cleaning and preprocessing, specifically including: S301. Wavelet threshold denoising algorithm is used to denoise voltage and current signals. The decomposition level is 3-5 levels. The db4 or sym5 wavelet basis is selected. S302. Use the Hampel identifier or 3σ criterion to identify and eliminate abnormal outliers caused by poor sensor contact or electromagnetic interference. S303. For missing data, use K-nearest neighbor-based interpolation or linear interpolation to fill in the missing data, ensuring the continuity and equal intervals of the data sequence.
[0023] Based on standard time-series datasets, sensitive feature parameters of batteries under critical health conditions are extracted, specifically including: Polynomial fitting was performed on the standard charge-discharge voltage curve to calculate the first derivative of voltage with respect to capacity. and the first derivative of capacitance with respect to voltage Extract the amplitude, area, and half-width of the characteristic peak, and calculate the offset of the characteristic peak relative to the initial state. and ,in, , For the first Voltage values at each sampling point For the first The cumulative capacity value of each sampling point For capacity intervals; Based on pulse charge-discharge data, the ohmic internal resistance is identified online using the equivalent circuit model ECM combined with the recursive least squares method. and polarization resistance Calculate the relative growth rate of internal resistance The formula is: ; in, For the ohmic internal resistance currently identified online, This refers to the initial ohmic internal resistance specified by the battery at the factory. During the resting period after charging and discharging is cut off, voltage recovery curves are collected and fitted with exponential decay functions. Extracting the time constant As a characterization parameter for lithium-ion diffusion ability, among which, This is the steady-state voltage at the end of the resting period. For voltage recovery amplitude, This refers to the settling time.
[0024] A fast battery health estimation model based on multi-dimensional feature fusion is constructed, specifically including: Establish a regression model based on machine learning, and select support vector machine, random forest or long short-term memory network as the base learner; The extracted sensitive feature parameters are used as the input vector. Defined as: ; in, This is the peak value of the differential voltage. For differential capacity peak value, The internal resistance growth rate, The voltage relaxation time constant is The standard deviation of the temperature field; The actual SOH value calibrated in the laboratory will be used as the label. ; The hyperparameters of the model are optimized using grid search or Bayesian optimization algorithm, and cross-validation is used to prevent overfitting. By introducing a transfer learning mechanism, the model is pre-trained using historical full life cycle data of batteries of the same model and batch, and then fine-tuned using a small amount of measured data of the current battery, achieving high-precision estimation with small samples.
[0025] Output the current battery health assessment value and confidence interval, specifically including: An ensemble learning framework is used to run multiple different sub-models simultaneously, such as SVM, GPR, and XGBoost. Based on the historical performance of each sub-model on the validation set, dynamically allocate weight coefficients. ,satisfy The final result is obtained by weighted summation. The point estimate is calculated using the following formula: ; in, The number of sub-models, Let i be the predicted value of the i-th sub-model. For the corresponding weights; Calculate the variance or standard deviation of the prediction results of each sub-model, and combine it with the prediction variance of the Gaussian process regression to construct a confidence interval with a 95% confidence level. Specifically: ; in, The standard deviation of the predicted values; If the confidence interval width Greater than the preset threshold If the current data quality is insufficient, a re-collection command will be triggered.
[0026] Please refer to Figure 4 As shown, the logic for determining whether a battery has entered the critical health retirement period is as follows: S401, Define the critical retirement period for health as It is in the 80%-85% range and the rate of decline is accelerating; S402. Set a dual threshold determination mechanism. First threshold (hard threshold): when If ≤80%, it will be directly judged as retirement; Second threshold (soft threshold / early warning threshold): When 80% < When ≤85%, calculate the most recent A cycle attenuation slope The calculation formula is: ; S403, if the attenuation slope Greater than the historical average decay rate of times, and ≥1.5, that is, it satisfies > And the internal resistance growth rate If it exceeds 30%, it is determined to enter the "critical warning state"; S404. Introduce a hysteresis comparison mechanism to prevent... Frequent alarms caused by fluctuations around the threshold, hysteresis width Set it to 1%-2%.
[0027] Please refer to Figure 5 As shown, it also includes a trend prediction step based on historical trajectories: S501, retrieve historical charge / discharge cycle data stored in the battery management system, and establish... The degradation trajectory varies with the number of cycles or mileage; S502. Using a sliding time window, such as the last 50 cycles, perform linear or nonlinear fitting on the degradation trajectory and calculate the current degradation acceleration. S503, combined with the Arrhenius equation, based on the current average operating temperature And depth of charge / discharge, predicting the short-term future Evolution trend, decay rate constant Expressed as: ; in, Pre-exponential factor, For activation energy, For molar gas constant, This refers to the average operating temperature. S504, if predicted If the value is about to fall below the retirement threshold within a short period, an "imminent retirement" warning will be triggered ahead of time.
[0028] It also includes cloud-based big data-assisted correction steps: The current battery's feature parameter vector is uploaded to the cloud server and compared with the full life cycle data of batteries of the same model and production batch in the cloud database; Using dynamic time warping The algorithm calculates the similarity between the current battery degradation curve and the standard retirement curve. The calculation formula is: ; in, This is the current battery feature sequence. For standard battery characteristic sequences, To standardize the path, The distance is Euclidean. If the similarity is higher than the preset value, such as 0.9, the remaining lifetime corresponding to the standard curve is directly used as the reference value to correct the prediction results of the local model and realize multi-source data fusion diagnosis.
[0029] In a second aspect of the invention, a rapid detection system for the critical health retirement period of a power lithium battery is also provided, comprising: The acquisition module is used to acquire multi-dimensional state data of the battery in real time during the charging and discharging process at a preset high-frequency sampling period. The multi-dimensional state data includes terminal voltage, charging and discharging current, battery surface temperature, internal impedance spectrum and single cell voltage difference. The preprocessing module is used to clean and preprocess the collected raw data, remove abnormal jump points and align timestamps to generate a standard time series dataset. The extraction module is used to extract sensitive characteristic parameters of the battery under critical health conditions based on a standard time series dataset. The sensitive characteristic parameters include the peak slope of the differential voltage curve, the peak offset of the differential capacity curve, the growth rate of the ohmic internal resistance, and the voltage relaxation characteristic time constant. The output module is used to build a fast battery health estimation model based on multi-dimensional feature fusion. It inputs the extracted sensitive feature parameters into the model and outputs the current battery health assessment value and confidence interval. The determination module is used to determine whether the battery has entered the critical retirement period based on the health assessment value and its rate of change, combined with a preset retirement threshold window, and outputs an alarm signal or retirement suggestion.
[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. A rapid detection method for the critical retirement period of a power lithium battery, characterized in that, include: During the charging and discharging process of the battery, multi-dimensional state data of the battery are collected in real time at a preset high-frequency sampling period. The multi-dimensional state data includes terminal voltage, charging and discharging current, battery surface temperature, internal impedance spectrum and single cell voltage difference. The collected raw data is cleaned and preprocessed, abnormal jump points are removed and timestamps are aligned to generate a standard time series dataset; Based on the standard time series dataset, sensitive characteristic parameters of the battery under critical health conditions are extracted. These sensitive characteristic parameters include the peak slope of the differential voltage curve, the peak offset of the differential capacity curve, the ohmic internal resistance growth rate, and the voltage relaxation characteristic time constant. A fast battery health estimation model based on multi-dimensional feature fusion is constructed. The extracted sensitive feature parameters are input into the model, and the current battery health assessment value and confidence interval are output. Based on the health assessment value and its rate of change, combined with the preset retirement threshold window, it is determined whether the battery has entered the critical retirement period and an alarm signal or retirement suggestion is output.
2. The rapid detection method for the critical retirement period of a power lithium battery according to claim 1, characterized in that, The real-time acquisition of multi-dimensional state data of the battery at a preset high-frequency sampling period specifically includes: Data acquisition is performed when the battery is in dynamic operating conditions, such as DST or FUDS conditions, or under short-term high-rate charge and discharge pulse excitation. A sliding window mechanism is adopted, with a window length of T and 10s≤T≤60s. Within the window, transient response data of voltage and current are collected at a frequency of milliseconds. Simultaneously, thermal imaging data of the battery module is collected, and the temperature field distribution gradient on the battery surface is obtained through infrared thermometry, serving as an auxiliary data source for thermal runaway early warning.
3. The rapid detection method for the critical retirement period of a power lithium battery according to claim 2, characterized in that, The cleaning and preprocessing of the collected raw data specifically includes: The voltage and current signals are denoised using a wavelet threshold denoising algorithm with 3-5 decomposition layers, and the db4 or sym5 wavelet basis is selected. Use Hampel identifiers or the 3σ criterion to identify and eliminate outliers caused by poor sensor contact or electromagnetic interference. For missing data, K-nearest neighbor-based interpolation or linear interpolation is used to fill in the missing data to ensure the continuity and equal spacing of the data sequence.
4. The rapid detection method for the critical retirement period of a power lithium battery according to claim 3, characterized in that, The extraction of sensitive feature parameters of the battery under critical health conditions based on the standard time-series dataset specifically includes: Polynomial fitting was performed on the standard charge-discharge voltage curve to calculate the first derivative of voltage with respect to capacity. and the first derivative of capacitance with respect to voltage Extract the amplitude, area, and half-width of the characteristic peak, and calculate the offset of the characteristic peak relative to the initial state. and ,in, , For the first Voltage values at each sampling point For the first The cumulative capacity value of each sampling point For capacity intervals; Based on pulse charge-discharge data, the ohmic internal resistance is identified online using the equivalent circuit model ECM combined with the recursive least squares method. and polarization resistance Calculate the relative growth rate of internal resistance The formula is: ； in, For the ohmic internal resistance currently identified online, This refers to the initial ohmic internal resistance specified by the battery at the factory. During the resting period after charging and discharging is cut off, voltage recovery curves are collected and fitted with exponential decay functions. Extracting the time constant As a characterization parameter for lithium-ion diffusion ability, among which, This is the steady-state voltage at the end of the resting period. For voltage recovery amplitude, This refers to the settling time.
5. The rapid detection method for the critical retirement period of a power lithium battery according to claim 4, characterized in that, The construction of a fast battery health estimation model based on multi-dimensional feature fusion specifically includes: Establish a regression model based on machine learning, and select support vector machine, random forest or long short-term memory network as the base learner; The extracted sensitive feature parameters are used as the input vector. Defined as: ； in, This is the peak value of the differential voltage. For differential capacity peak value, The internal resistance growth rate, The voltage relaxation time constant is The standard deviation of the temperature field; The actual SOH value calibrated in the laboratory will be used as the label. ; The hyperparameters of the model are optimized using grid search or Bayesian optimization algorithm, and cross-validation is used to prevent overfitting. By introducing a transfer learning mechanism, the model is pre-trained using historical full life cycle data of batteries of the same model and batch, and then fine-tuned using a small amount of measured data of the current battery, achieving high-precision estimation with small samples.
6. The rapid detection method for the critical retirement period of a power lithium battery according to claim 5, characterized in that, The output of the current battery health assessment value and confidence interval specifically includes: An ensemble learning framework is used to run multiple different sub-models simultaneously, such as SVM, GPR, and XGBoost. Based on the historical performance of each sub-model on the validation set, dynamically allocate weight coefficients. ,satisfy The final result is obtained by weighted summation. The point estimate is calculated using the following formula: ； in, The number of sub-models, Let i be the predicted value of the i-th sub-model. For the corresponding weights; Calculate the variance or standard deviation of the prediction results of each sub-model, and combine it with the prediction variance of the Gaussian process regression to construct a confidence interval with a 95% confidence level. Specifically: ； in, The standard deviation of the predicted values; If the confidence interval width Greater than the preset threshold If the current data quality is insufficient, a re-collection command will be triggered.
7. A rapid detection method for the critical retirement period of a power lithium battery according to claim 6, characterized in that, The logic for determining whether a battery has entered a critical health retirement period is as follows: Define the critical retirement period for health as It is in the 80%-85% range and the rate of decline is accelerating; A dual threshold determination mechanism is set up, the First threshold (hard threshold): when If the percentage is ≤80%, the employee is directly deemed to be retired. Second threshold (soft threshold / early warning threshold): When 80% < When ≤85%, calculate the most recent A cycle attenuation slope The calculation formula is: ； If the attenuation slope Greater than the historical average decay rate of times, and ≥1.5, that is, it satisfies > And the internal resistance growth rate If it exceeds 30%, it is determined to enter the "critical warning state"; Introduce a hysteresis comparison mechanism to prevent Frequent alarms caused by fluctuations around the threshold, hysteresis width Set it to 1%-2%.
8. The rapid detection method for the critical retirement period of a power lithium battery according to claim 7, characterized in that, It also includes a trend prediction step based on historical trajectories: Retrieve historical charge-discharge cycle data stored in the battery management system and establish The degradation trajectory varies with the number of cycles or mileage; Using a sliding time window, such as the last 50 cycles, linear or nonlinear fitting is performed on the degradation trajectory to calculate the current degradation acceleration; Combining the Arrhenius equation, based on the current average operating temperature And depth of charge / discharge, predicting the short-term future Evolution trend, decay rate constant Expressed as: ； in, Pre-exponential factor, For activation energy, For molar gas constant, This refers to the average operating temperature. If the prediction If the value is about to fall below the retirement threshold within a short period, an "imminent retirement" warning will be triggered ahead of time.
9. A rapid detection method for the critical retirement period of a power lithium battery according to claim 8, characterized in that, It also includes cloud-based big data-assisted correction steps: The current battery's feature parameter vector is uploaded to the cloud server and compared with the full life cycle data of batteries of the same model and production batch in the cloud database; Using dynamic time warping The algorithm calculates the similarity between the current battery degradation curve and the standard retirement curve. The calculation formula is: ； in, This is the current battery feature sequence. For standard battery characteristic sequences, To standardize the path, The distance is Euclidean. If the similarity is higher than the preset value, such as 0.9, the remaining lifetime corresponding to the standard curve is directly used as the reference value to correct the prediction results of the local model and realize multi-source data fusion diagnosis.
10. A rapid detection system for the critical retirement period of a power lithium battery, used to implement the rapid detection method for the critical retirement period of a power lithium battery as described in any one of claims 1-9, characterized in that, include: The acquisition module is used to acquire multi-dimensional state data of the battery in real time during the charging and discharging process at a preset high-frequency sampling period. The multi-dimensional state data includes terminal voltage, charging and discharging current, battery surface temperature, internal impedance spectrum and single cell voltage difference. The preprocessing module is used to clean and preprocess the collected raw data, remove abnormal jump points and align the timestamps to generate a standard time series dataset. The extraction module is used to extract sensitive characteristic parameters of the battery under critical health conditions based on the standard time series dataset. The sensitive characteristic parameters include the peak slope of the differential voltage curve, the peak offset of the differential capacity curve, the ohmic internal resistance growth rate, and the voltage relaxation characteristic time constant. The output module is used to construct a fast battery health estimation model based on multi-dimensional feature fusion, input the extracted sensitive feature parameters into the model, and output the current battery health assessment value and confidence interval. The determination module is used to determine whether the battery has entered the critical retirement period based on the health assessment value and its rate of change, combined with a preset retirement threshold window, and output an alarm signal or retirement suggestion.