Remaining life prediction method for retired power cell

Abstract

The invention provides a method for predicting the residual life of a decommissioned power battery cell, and the method comprises the steps: carrying out the impedance testing of different frequencies of a to-be-tested battery cell in a decommissioned power battery in a preset testing frequency range, obtaining the impedance response data at different frequencies, and obtaining the impedance spectrum data; utilizing scene conditions corresponding to different steps are obtained, the utilizing scene conditions comprise the temperature range and the charging and discharging multiplying power, and a target step utilizing scene is matched in combination with the step grade to which the current to-be-detected battery cell belongs; performing associated storage on the target step utilization scene and the model, capacity, internal resistance and impedance spectrum data information of the retired battery cell to form a retired battery step utilization scheme; if it is detected that the battery is in an abnormal state, an early warning mechanism is triggered, and corresponding safety protection measures including cutting off a power source and starting a cooling system are taken according to a preset processing strategy.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for predicting the remaining life of a retired power battery cell. Background Art

[0002] In the scenario of step-by-step utilization of retired power batteries, it is necessary to evaluate the remaining life of the battery cells in order to make reasonable step-by-step divisions. First of all, how to obtain the impedance spectrum data of the battery cells in retired power batteries is a key step in predicting the remaining life of the battery cells. However, impedance spectrum data usually contains multiple characteristic parameters, such as ohmic internal resistance, polarization resistance, diffusion resistance, etc. Different characteristic parameters have different degrees of influence on the remaining life of the battery cells, and it is necessary to select appropriate characteristic parameters as the input of the life prediction model. At the same time, during the acquisition of impedance spectrum data, the selection of measurement frequency will also affect the quality of the data and the accuracy of the subsequent model. Therefore, how to obtain high-quality impedance spectrum data under complex working conditions and extract effective characteristic parameters from it is a key technical problem faced by the remaining life assessment of retired power batteries. Furthermore, how to divide the steps according to the remaining life is also a major technical challenge in this technical field, and matching the corresponding utilization scenarios according to the divided steps is also a technical difficulty to be solved. In addition, in the actual step-by-step utilization process, how to ensure the safety and reliability of utilization is also one of the technical issues that need to be considered in this solution. Summary of the invention

[0003] The present invention provides a method for predicting the remaining life of a retired power cell, which mainly includes:

[0004] Obtain the basic parameter information of the battery cell to be tested in the retired power battery, including the battery cell model, capacity, and internal resistance, and obtain the standard impedance spectrum data of the battery cell of this model from the preset battery cell parameter database according to the battery cell model;

[0005] For the cells to be tested in the retired power batteries, impedance tests at different frequencies are performed within a preset test frequency range to obtain impedance response data at different frequencies and obtain impedance spectrum data;

[0006] Preprocess the acquired impedance spectrum data, including removing abnormal points, smoothing data, extracting key characteristic parameters of the impedance spectrum curve, including arc radius, center coordinates, slope, and constructing impedance spectrum feature vectors;

[0007] The impedance spectrum feature vector is input into the pre-built life prediction model, and the predicted remaining life data of the battery cell to be tested is output. Combined with the preset step utilization division rule, the step level to which the battery cell to be tested belongs is obtained;

[0008] Obtain the usage scenario conditions corresponding to different levels, including temperature range and charge / discharge rate, and match the target level usage scenario in combination with the level to which the current battery cell to be tested belongs;

[0009] The target step-by-step utilization scenario is associated with the retired battery cell model, capacity, internal resistance and impedance spectrum data information to form a step-by-step utilization plan for retired batteries;

[0010] In the actual step-by-step utilization process, according to the formed step-by-step utilization plan for retired batteries, the real-time working data of retired power batteries in the corresponding utilization scenarios, including voltage, current, and temperature, are collected, and the preset abnormality detection model is used to determine whether the battery is in an abnormal state;

[0011] If an abnormal battery condition is detected, the early warning mechanism will be triggered and corresponding safety protection measures will be taken according to the preset processing strategy, including cutting off the power supply and starting the cooling system.

[0012] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0013] The present invention discloses a method for predicting the remaining life of retired power batteries. The method obtains the basic parameters and impedance spectrum data of retired battery cells, combines a preset life prediction model and a step-by-step utilization division rule, and realizes accurate evaluation of the remaining life of retired batteries and the division of step levels. Furthermore, the present invention matches the corresponding utilization scenario according to the step level to which the battery belongs, and forms a step-by-step utilization plan for retired batteries. In the actual utilization process, the present invention also monitors the battery working data in real time, combines the abnormal detection model, and timely identifies the abnormal state of the battery and triggers the corresponding safety protection measures. This method not only improves the utilization efficiency of retired power batteries, but also ensures the safety of the step-by-step utilization process, providing an effective solution for the full life cycle management of power batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a flowchart of a method for predicting the remaining life of retired power cells.

[0015] Figure 2 It is a schematic diagram of a method for predicting the remaining life of retired power batteries according to the present invention.

[0016] Figure 3 It is another schematic diagram of a method for predicting the remaining life of retired power cells according to the present invention. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0018] like Figure 1-3 In this embodiment, a method for predicting the remaining life of a retired power cell may specifically include:

[0019] S101. Obtain basic parameter information of the battery cell to be tested in the retired power battery, including the battery cell model, capacity, and internal resistance, and obtain standard impedance spectrum data of the battery cell of this model from a preset battery cell parameter database according to the battery cell model.

[0020] According to the identification of retired power cells, design specification parameters are obtained from a preset database, where the design specification parameters include a nominal capacity value, a rated voltage value and a standard internal resistance value; the retired power cells are charged to the rated voltage value using a constant current and constant voltage method, and a stable state cell is obtained by standing for a preset time; an electrochemical impedance spectrometer is used to perform sweep frequency measurement on the stable state cell to obtain raw impedance data, and a real and imaginary impedance numerical sequence is generated through fast Fourier transform; the impedance numerical sequence is verified for data validity, abnormal data points are eliminated, and missing data are supplemented using cubic spline interpolation to generate a continuous impedance curve.

[0021] Specifically, the design specification parameters are read from the preset database according to the retired power battery cell identification, the battery cell model code and production batch number are matched with the data table, and the nominal capacity value, rated voltage value, and standard internal resistance value recorded when the battery cell of this model leaves the factory are obtained from the parameter database. The ambient temperature and humidity are controlled for the battery cell to be tested, so that the test environment temperature is kept at a constant temperature of 25 degrees Celsius, the battery cell is charged to the rated voltage value using a constant current and constant voltage method, and the battery cell is left to stand for a preset time to reach a stable state. An electrochemical impedance spectrometer is used to perform sweep frequency measurement on the battery cell within a preset frequency range to obtain the original data of the battery cell impedance, and the measured data is subjected to spectrum analysis through fast Fourier transform to generate a real and imaginary impedance numerical sequence. The data validity of the impedance numerical sequence is verified, abnormal data points are eliminated, and the missing data is supplemented using cubic spline interpolation to generate a continuous impedance curve, and the charge transfer impedance point and double layer capacitance point positions are marked on the spectrum graph. The measured cell impedance curve is compared with the standard impedance curve of the cell of this model, and the relative offset of the feature point position is calculated. If the offset exceeds the preset threshold range, the pre-trained convolution model is used to fit the impedance spectrum data. The least squares method is used to calculate the matching difference rate between the measured impedance curve and the standard impedance curve, and the difference rate data is normalized by the hyperbolic tangent function to generate a standardized feature vector. The standardized feature vector is input into the pre-trained support vector regression model, which uses a Gaussian kernel function. The input parameters include the impedance feature point offset, the internal resistance difference rate, and the voltage response curve characteristic value, and the output cell capacity attenuation degree value. The performance of retired power cells will deteriorate due to cyclic charging and discharging and environmental factors during use. The changes in the internal impedance characteristics of the cell can be obtained by electrochemical impedance spectroscopy measurement. Taking lithium iron phosphate cells as an example, the nominal capacity is 60 ampere hours, the rated voltage is 3.2 volts, and the standard internal resistance value at the factory is 0.8 milliohms. After the battery cell has been used for a certain period of time, the measured internal resistance value may rise to 1.2 milliohms, indicating that performance degradation has occurred inside the battery cell. Before performing impedance measurement, the battery cell needs to be charged and pre-treated to put the battery cell in a suitable test state. For lithium iron phosphate batteries, 0.5 times the rated current is used for constant current charging to 3.2 volts, and then maintained at a constant voltage for 2 hours to stabilize the electrochemical reaction inside the battery cell. The test environment temperature is maintained at 25 degrees Celsius and the relative humidity is controlled at 45% to avoid environmental factors from interfering with the measurement results. The measurement frequency range of the impedance spectrum is set between 0.01 Hz and 100 kHz, and impedance data at different frequencies is obtained by logarithmic frequency sweeping. After fast Fourier transform processing, typical semicircular impedance spectrum characteristics can be observed on the Nyquist plot. Among them, the high-frequency region reflects the resistance characteristics of the electrolyte, the semicircle diameter of the medium-frequency region corresponds to the charge transfer impedance, and the low-frequency region reflects the diffusion characteristics of lithium ions in the electrode material.The 3-times standard deviation principle is used to process abnormal data, and data points that deviate from the mean by more than 3 times the standard deviation are judged as outliers and removed. For example, the impedance value measured at a frequency of 50 Hz is 2.5 milliohms, while the mean at this frequency is 1.5 milliohms and the standard deviation is 0.2 milliohms. Since the measured value deviates from the mean by more than 3 times the standard deviation, it is removed and the data is supplemented by the cubic spline interpolation method. The offset of the characteristic points of the impedance curve reflects the degree of deterioration of different components inside the battery cell. When the charge transfer impedance point shifts to the right by more than 30%, it indicates that the activity of the electrode material is reduced. By analyzing the changing trends of parameters such as charge transfer impedance and double-layer capacitance, the capacity attenuation of the battery cell can be evaluated in combination with the support vector regression model. When the capacity attenuation exceeds 20%, the battery cell is no longer suitable for continued use in the field of power batteries, but can be downgraded for use in fields such as energy storage. This evaluation method based on impedance characteristics provides an important basis for the hierarchical utilization of retired power batteries.

[0022] S102, performing impedance tests at different frequencies within a preset test frequency range on the cells to be tested in the retired power batteries, acquiring impedance response data at different frequencies, and obtaining impedance spectrum data.

[0023] The retired power cells are charged with constant current and constant voltage until the charging cut-off voltage value, and the charged cells are discharged with rated current using a constant current power supply until the discharge cut-off voltage value; the open circuit voltage of the discharged cells is monitored, and the cells are determined to be in a stable state based on the open circuit voltage fluctuation value being less than a preset voltage fluctuation threshold; a sinusoidal AC excitation signal is applied to the cells in a stable state using an electrochemical impedance spectrometer, and the excitation signal response is subjected to Fourier spectrum conversion to obtain an impedance numerical sequence; the impedance numerical sequence is supplemented with points using a cubic spline interpolation method to obtain an impedance data curve, and the positions of the semicircular arc feature points are identified according to the impedance data curve to obtain an impedance spectrum feature data set.

[0024] Specifically, the retired power cell is charged with constant current and constant voltage according to the rated voltage value of the cell, and the surface temperature value of the cell is collected through a temperature sensor. When the temperature measurement value is in the range of 20 to 30 degrees Celsius, a constant current power supply is used to apply 0.2 times the rated current to the cell for constant current discharge until the cell voltage value reaches the discharge cut-off voltage value. The cell is left to stand, and the open circuit voltage of the cell is continuously monitored by the voltage acquisition module. When the open circuit voltage fluctuation value is less than a preset threshold, the cell is determined to be in a stable state. An electrochemical impedance spectrometer is used to apply a sinusoidal AC excitation signal with an amplitude of 5% of the charging current to the stable state cell. Starting from the kilohertz frequency starting point, the frequency is swept point by point to the millihertz level in a manner of reducing one order of magnitude every ten frequency points, and the voltage response signal and current excitation signal at each frequency point are recorded by a high-precision data collector. The signal-to-noise ratio is calculated for the obtained response signal, the frequency point data with a signal-to-noise ratio lower than the preset threshold is eliminated, the remaining response signal is Fourier spectrum transformed, the voltage amplitude and phase value of each frequency sampling point are calculated, and the real and imaginary values ​​of the impedance are obtained by complex number operation. The frequency intervals in the impedance numerical sequence are encrypted and supplemented by cubic spline interpolation to generate a continuous impedance data curve, and the curve is subjected to noise reduction and smoothing by wavelet multi-scale decomposition. Feature extraction is performed on the smoothed impedance data curve, the positions of the starting point, vertex, and end point of the semicircular arc are identified, the frequency value and impedance value corresponding to each feature point are recorded, and a complete impedance spectrum feature data set is generated. Electrochemical impedance spectroscopy measurement has high control requirements on the state of the battery cell, and the battery cell needs to be pre-processed before measurement. Taking the ternary lithium battery as an example, when the rated voltage is 3.7 volts, 0.2 times the rated current is used for constant current charging to 4.2 volts, and then 4.2 volts are used for constant voltage charging until the charging current drops to 0.05 times the rated current, at which time the battery cell reaches a full charge state. During the charging process, the surface temperature of the battery cell rises and needs to be left to stand until the temperature drops to around 25 degrees Celsius. During the static state of the battery cell, the open circuit voltage will fluctuate slightly. When the voltage fluctuation value is less than 2 millivolts within 30 minutes of continuous monitoring, it indicates that the electrochemical reaction inside the battery cell tends to balance. The open circuit voltage measured at this time is 4.186 volts, which is within a reasonable range from the theoretical value of 4.2 volts, indicating that the state of the battery cell is suitable for impedance measurement. The impedance measurement adopts a small signal excitation method, and the frequency is swept with an AC signal of 5% of the rated charging current. For a 60 ampere-hour battery, the excitation current amplitude is 3 amperes. The frequency sweep range starts at 1000 Hz and decreases by one order of magnitude every 10 frequency points until 0.01 Hz, with a total of 61 frequency sampling points. 16 cycles of data are collected at each frequency point. The acquisition time of a single frequency point in the high frequency band is short, while a single point acquisition at 0.01 Hz requires 1600 seconds. There is noise interference in the original acquisition signal. Data screening is performed by calculating the signal-to-noise ratio. When the signal-to-noise ratio is lower than 20 decibels, it is judged as invalid data.The impedance data obtained by Fourier transform showed that the impedance value was 0.8 milliohms at 1000 Hz, which mainly reflected the resistance of the electrolyte; a semicircular arc was formed in the range of 100 Hz to 1 Hz, corresponding to the charge transfer process, and the diameter of the semicircular arc was 1.2 milliohms; below 0.1 Hz, it appeared as a slant line, reflecting the diffusion characteristics of lithium ions in the electrode material. The impedance curve was interpolated with cubic spline, and 9 data points were added at each interval on the basis of the original 61 frequency points to obtain a continuous curve of 541 points. The db4 wavelet was used to decompose the curve into 4 layers, and the high-frequency noise was removed and reconstructed to obtain a smooth curve. On the Nyquist diagram, the point with the minimum value of the real resistance was the starting point of the semicircular arc, and the point with the maximum value of the real resistance was the ending point. The position of the vertex of the semicircular arc was determined by curvature calculation. The position and value of these characteristic points reflect the dynamic characteristics of different physical processes inside the battery and are important indicators for evaluating battery performance.

[0025] S103, preprocessing the acquired impedance spectrum data, including removing abnormal points, smoothing data, extracting key characteristic parameters of the impedance spectrum curve, including arc radius, center coordinates, and slope, and constructing an impedance spectrum characteristic vector.

[0026] A sliding window is used to calculate the local variance value of the impedance spectrum data sequence. If the variance value exceeds the mean threshold of all variances in the frequency interval, the marked abnormal data points are removed from the impedance spectrum data sequence. The interval value between adjacent frequency points is calculated according to the impedance spectrum data sequence after the abnormal points are removed. If the interval exceeds the preset frequency step, a linear interpolation method is used to obtain supplementary frequency point data. For the impedance spectrum sequence after the supplementary frequency point data, a Butterworth low-pass filter and a four-layer discrete wavelet decomposition are used to remove high-frequency noise components, and a continuous and smooth impedance spectrum curve is obtained by a cubic spline function. According to the continuous and smooth impedance spectrum curve, arc fitting and linear fitting are performed on the complex plane, and the iterative least squares method is used to obtain the center coordinate value and radius value. The real part value and imaginary part value of the low-frequency slope are obtained by complex number operation, and the eigenvector is normalized.

[0027] Specifically, according to the fluctuation amplitude between adjacent data points in the original data sequence of the impedance spectrum, a sliding window with a length of five data points is used to calculate the local variance value. If the variance value exceeds three times the mean value of all variances in the frequency interval, the data point is marked as an abnormal point, and the marked abnormal data point is removed from the impedance spectrum data sequence. The data integrity of the impedance spectrum data sequence after the abnormal points are removed is verified, and the interval value between adjacent frequency points is calculated. If the interval exceeds the preset frequency step, the missing frequency point data is supplemented by a linear interpolation method. For the impedance spectrum sequence after the supplemented data, a Butterworth low-pass filter with a cutoff frequency of one twentieth of the sampling frequency is used for processing, and the high-frequency noise component is removed by four-layer discrete wavelet decomposition. The data points are fitted with a cubic spline function to obtain a continuous and smooth impedance spectrum curve. The impedance spectrum data with a frequency range of ten hertz to kilohertz are fitted with an arc on the complex plane, and the coordinate value and radius value of the center of the arc are calculated by the iterative least squares method. The iteration is stopped when the fitting residual is less than the preset threshold. The impedance spectrum data with a frequency less than ten hertz is linearly fitted by the least squares method, and the slope value of the impedance curve in the low-frequency region is calculated, and the component values ​​of the slope in the real part and the imaginary part are obtained by complex number operations. The horizontal coordinate value of the arc center, the vertical coordinate value of the center, the arc radius value, the real part value of the low-frequency slope, and the imaginary part value of the low-frequency slope are combined in a preset order to construct a five-dimensional feature vector, and the physical meaning of each feature component is recorded. The principal component analysis method is used to reduce the dimension of the five-dimensional feature vector, and the principal component with a cumulative contribution rate exceeding the preset threshold is selected as the new feature vector, and the feature vector is normalized by calculating the Mahalanobis distance from the data point to the center of mass. The preprocessing of impedance spectrum data is a key link in extracting effective features, and the original data often contains various anomalies and noise. In actual measurements, taking a 60 ampere-hour ternary lithium battery as an example, the impedance value measured at a frequency of 1000 Hz is 0.8 milliohms, while the impedance values ​​at the adjacent 999 Hz and 1001 Hz frequency points are 0.79 milliohms and 2.1 milliohms, respectively. It is obvious that the data at 1001 Hz is abnormal. A 5-point sliding window is used to calculate the variance. The local variance at the abnormal point is 0.42, which is far beyond the average variance value of 0.005 in the frequency band, so it is marked as an abnormal point and removed. Data integrity verification shows that there is a large frequency interval in the range of 100 Hz to 10 Hz, and the interval between adjacent frequency points reaches 15 Hz, which exceeds the preset 10 Hz step requirement. Data at intermediate frequency points such as 12 Hz and 14 Hz are added by linear interpolation to make the frequency distribution more uniform. The supplemented data is processed by a 4th-order Butterworth low-pass filter with a cutoff frequency set to 50 Hz, which can effectively remove high-frequency interference during the measurement process. On the complex plane, the impedance data in the range of 100 Hz to 1 Hz presents a typical semicircular arc shape, and the iterative least squares method is used for arc fitting.The initial center coordinates were set to 1.2 milliohms for the real part, 0.6 milliohms for the imaginary part, and 0.8 milliohms for the radius. After 20 iterations, the fitting residual dropped below 0.01, and the final center coordinates were 1.25 milliohms for the real part, 0.62 milliohms for the imaginary part, and 0.82 milliohms for the radius. In the low-frequency region where the frequency is less than 1 Hz, the impedance curve presents a slope of about 45 degrees, reflecting the characteristics of the diffusion process. The real part of the slope was 0.71 and the imaginary part was 0.68 by least squares fitting, which is close to the 45-degree angle of the ideal diffusion process. The arc features and low-frequency features are combined into a five-dimensional vector [1.25, 0.62, 0.82, 0.71, 0.68], which is reduced to three dimensions through principal component analysis, and the cumulative contribution rate reaches 95%. These characteristic components reflect the characteristics of electrolyte resistance, charge transfer resistance, and diffusion impedance, respectively, and are important indicators for evaluating battery performance. The Mahalanobis distance of the feature vectors of all samples is calculated and normalized to obtain standardized feature vectors, which is convenient for subsequent battery performance evaluation.

[0028] S104, inputting the impedance spectrum feature vector into a pre-built life prediction model, outputting the predicted remaining life data of the battery cell to be tested, and combining it with a preset step utilization division rule to obtain the step level to which the battery cell to be tested belongs.

[0029] The mean and standard deviation of each characteristic component are calculated according to the characteristic vector of the impedance spectrum, and the standardized characteristic vector is obtained by using the Gaussian distribution normalization method; the standardized characteristic vector is subjected to singular value decomposition, and the characteristic components whose cumulative contribution rate exceeds the contribution rate threshold are sorted according to the size of the singular values, and a reduced-dimensionality characteristic vector is obtained; the reduced-dimensionality characteristic vector is input into a preset deep neural network, and the output values ​​of the nodes at each layer are subjected to nonlinear transformation to obtain the predicted value of the remaining life of the battery cell; the fuzzy clustering method is used to calculate the membership of the predicted value of the remaining life of the battery cell to each level, and the grading result is verified by calculating the Mahalanobis distance between the characteristic vector of the battery cell and the center of the samples of each level to obtain the final grading data of the battery cell.

[0030] Specifically, the impedance spectrum feature vector is normalized and preprocessed, the feature data is scaled by Gaussian distribution standardization method, the mean and standard deviation of each feature component are calculated, and a standardized feature vector is generated. The standardized feature vector is denoised by singular value decomposition, sorted by singular value size, and the feature components whose cumulative contribution rate exceeds the preset threshold are retained to generate a feature vector after dimensionality reduction. The training weight parameters are read from a preset five-layer deep neural network, which includes three hidden layers, and the number of hidden layer nodes is sixteen, eight, and four, respectively. The input layer corresponds to the dimension of the feature vector after dimensionality reduction, and the output layer corresponds to the remaining life value of a single node. The reduced dimensionality feature vector is input into the deep neural network, and the output value of each layer node is calculated in turn by the forward propagation algorithm. The nonlinear transformation is performed by the sigmoid activation function to obtain the predicted value of the remaining life of the battery cell. The prediction result is verified by the five-fold cross-validation method, the mean and variance of the five prediction results are calculated, and the confidence of the prediction result is evaluated by the root mean square error. If the confidence is higher than the preset threshold, the prediction result is recorded. According to the recorded predicted values ​​of the remaining life of the battery cell, combined with the preset four-level ladder utilization grading standard, the fuzzy clustering method is used to calculate the membership of the predicted value to each level, and the level with the highest membership is the ladder level to which the battery cell belongs. The grading results are verified, and the Mahalanobis distance between the characteristic vector of the battery cell and the center of each level of samples is calculated. The level with the smallest distance should be consistent with the fuzzy clustering result to generate the final grading data of the battery cell. The life prediction of retired power batteries starts with the data preprocessing of the impedance characteristic vector. Taking a certain model of ternary lithium battery as an example, its impedance characteristic vector contains five characteristic components such as arc radius, center coordinates, and low-frequency slope. The dimensions and numerical ranges of the original characteristic data are quite different. For example, the arc radius is 0.82 milliohms, while the low-frequency slope is 45.3 degrees. Through Gaussian distribution standardization, the calculated feature means are 1.25, 0.62, 0.82, 0.71, and 0.68, and the standard deviations are 0.15, 0.08, 0.11, 0.09, and 0.08, respectively. The original data is converted into a standard normal distribution with a mean of 0 and a standard deviation of 1. Singular value decomposition is performed on the standardized feature vector, and 5 singular values ​​are obtained, which are 2.85, 1.42, 0.76, 0.35, and 0.12. The feature vectors corresponding to the first three singular values ​​with a cumulative contribution rate of 95% are selected to achieve dimensionality reduction and noise reduction. The deep neural network adopts a three-layer hidden layer structure. The input layer corresponds to a 3-dimensional feature vector, the first hidden layer has 16 nodes, the second hidden layer has 8 nodes, the third hidden layer has 4 nodes, and the output layer has 1 node corresponding to the remaining number of cycles. In the trained neural network, the standardized feature vector [-0.56, 0.82, 0.33] of a battery cell is input. After calculation of each layer of nodes and processing of the sigmoid activation function, the predicted number of remaining cycles is 453 times.Through five-fold cross validation, the samples were randomly divided into 5 parts, 4 parts were selected as training sets each time, and 1 part was selected as validation set. The 5 prediction results were 453, 448, 460, 442, and 456 times, with an average of 452 times and a standard deviation of 6.8 times. The calculated root mean square error was 1.5%, which was lower than the preset threshold of 2%, indicating that the prediction results had a high degree of confidence. According to the preset step utilization classification standard, the remaining number of cycles exceeding 800 times was divided into the first level, 500-800 times into the second level, 200-500 times into the third level, and less than 200 times into the fourth level. The membership of each level was calculated for the predicted value of 452 times, and the memberships were 0.05, 0.15, 0.75, and 0.05, respectively, and the battery cell was judged to belong to the third level. By calculating the Mahalanobis distance between the characteristic vector of the battery cell and the center of each level of samples, the distance values ​​are 8.6, 4.2, 1.8, and 5.3, respectively. The minimum distance corresponds to the third level, which is consistent with the fuzzy clustering result, verifying the reliability of the grading results. At this time, it can be determined that the battery cell is suitable for low-rate energy storage and other fields.

[0031] S105, obtaining utilization scenario conditions corresponding to different levels, the utilization scenario conditions including temperature range and charge / discharge rate, and matching the target level utilization scenario in combination with the level to which the current battery cell to be tested belongs.

[0032] A step-level utilization scenario table is read from a preset scenario parameter database, wherein the step-level utilization scenario table includes temperature range limits, charge and discharge rate limits, voltage range limits, cycle depth limits, and working time limits; a judgment matrix is ​​constructed according to the limit parameters in the step-level utilization scenario table, and weight coefficients of the limit parameters are calculated by a hierarchical analysis method; operating condition parameters of energy storage scenarios, power scenarios, and backup power scenarios are extracted based on the weight coefficients, and a scenario adaptability score is obtained by a data statistical method; the membership value of the battery cell to be tested to each scenario is calculated according to the scenario adaptability score, and a Monte Carlo method is used to verify whether the performance parameters of the battery cell to be tested meet the scenario requirements.

[0033] Specifically, a ladder-level utilization scenario table is read from the preset scenario parameter database, and the scenario table includes upper and lower limits of the temperature range, charge rate limit, discharge rate limit, upper and lower limits of the working voltage, cycle depth limit, and continuous working time limit. A judgment matrix is ​​constructed using the hierarchical analysis method, and the relative importance of five parameters, namely, temperature range, charge and discharge rate, voltage range, cycle depth, and working time, is calculated by pairwise comparison, and the eigenvalue and eigenvector are calculated to obtain the weight coefficient. The operating parameters are extracted for the energy storage scenario, power scenario, and backup power scenario, including the ambient temperature change curve, load power curve, charge and discharge timing, and working time distribution, and the characteristic parameter values ​​of each scenario are obtained using a data statistical method. According to the parameter limits of the ladder level to which the battery cell to be tested belongs, the difference with the characteristic parameters of each application scenario is calculated respectively, and the comprehensive difference score is calculated using the weighted summation method to generate a scenario adaptability score. The scenario adaptability score is normalized, and the fuzzy comprehensive evaluation method is used to calculate the membership value of the battery cell to be tested to each scenario to generate a scenario matching matrix. Based on the principle of maximum membership, the application scenario with the highest matching degree is selected, and the fluctuation analysis of the cell performance parameters in the scenario is performed by the Monte Carlo method to verify whether the cell parameters meet the scenario requirements. According to the results of the performance parameter fluctuation analysis, if the fluctuation range of the key parameters is within the scenario constraint range, the application scenario parameters are recorded as the cell cascade utilization plan to generate the recommended data for the cascade utilization scenario. The selection of cascade utilization scenarios for retired power batteries needs to consider the matching degree of multiple performance parameters and application conditions. Taking ternary lithium batteries as an example, strict parameter limits are set for different levels of cells in the cascade utilization scenario table. The operating temperature range of the first-level cell is -20 to 55 degrees Celsius, the charge rate limit is 1 times, the discharge rate limit is 2 times, the operating voltage range is 3.0 to 4.2 volts, the cycle depth limit is 80%, and the continuous working time limit is 4 hours. When determining the weight of the scenario parameters, the hierarchical analysis method is used to construct a 5-order judgment matrix. The importance of the temperature range is 2 compared with the charge and discharge rate, 3 compared with the voltage range, 4 compared with the cycle depth, and 5 compared with the working time. By calculating the maximum eigenvalue and eigenvector of the judgment matrix, the weights of the temperature range, charge and discharge rate, voltage range, cycle depth, and working time are obtained to be 0.42, 0.26, 0.16, 0.10, and 0.06, respectively. According to the working characteristics of the energy storage scenario, the ambient temperature fluctuates between 15 and 35 degrees Celsius, the load power curve shows two peaks in the morning and evening, the charge and discharge sequence is highly regular, and the single working time is 2 to 6 hours. The temperature in the power scenario changes dramatically, the load power fluctuates greatly, the charge and discharge are highly random, and the working time is not fixed. The temperature in the backup power scenario is relatively stable, the load power is stable, the number of charge and discharge times is small, and the single working time is short.The matching degree of the cell parameter limit and the scene characteristic parameters is calculated. The temperature adaptability score adopts the overlapping interval ratio, the charge and discharge rate score adopts the margin ratio, the voltage interval score adopts the coverage degree, and the cycle depth and working time score adopts the satisfaction rate. The calculation results of a certain three-level cell show that the temperature score of the energy storage scene is 0.85, the rate score is 0.92, the voltage score is 0.88, the depth score is 0.78, and the duration score is 0.82. After weighted summation, the comprehensive score of the energy storage scene is 0.86, the power scene is 0.65, and the backup power scene is 0.72. The Monte Carlo method is used to perform 1000 parameter fluctuation simulations to analyze the performance of the cell in the energy storage scenario. The results show that the temperature fluctuation is within the limit range under 95% of the working conditions, the charge and discharge rate meets the requirements under 98% of the working conditions, and the cycle depth does not exceed the limit under 93% of the working conditions. Comprehensive evaluation shows that the cell is suitable for energy storage scenarios, and the specific application direction is household energy storage or industrial and commercial energy storage.

[0034] S106: The target step-by-step utilization scenario is associated with the retired battery cell model, capacity, internal resistance and impedance spectrum data information to form a retired battery step-by-step utilization plan.

[0035] The model identification, capacity value, internal resistance value, and impedance spectrum data are concatenated to generate an original identification string, and a solution identification code is obtained through a secure hash algorithm; a parameter information table, an impedance spectrum data table, a level information table, and a scenario information table are established according to the solution identification code, and the solution identification code is used as the primary key to establish the association relationship between the data tables; a null value detection method is used to identify missing data items in the data table, and abnormal data markers are obtained through a data type verification method, and a data quality report is generated for the abnormal data markers; a structured query statement is used to extract complete data from the data table, and a retired battery step-by-step utilization plan document containing a basic information area, a feature data area, and a scenario parameter area is generated according to the complete data.

[0036] Specifically, according to the model identification, capacity value, internal resistance value, and impedance spectrum data of the retired battery cell, an original identification string is generated by information splicing, a 256-bit solution identification code is generated by a secure hash algorithm, and the current timestamp is recorded as the solution creation time. Four data tables, namely, parameter information table, impedance spectrum data table, level information table, and scenario information table, are established for the retired battery cell data. The solution identification code is used as the primary key to establish an association relationship, and secondary indexes are created for the solution time, battery cell model, and scenario type. The data items in the four data tables are checked for integrity, and the null value detection method is used to identify missing data items. The data format is checked by the data type verification method, and the data items that do not meet the specifications are marked. A data quality report is generated for the marked abnormal data items, and the location, type, and severity of the abnormal data are recorded. The missing data is repaired by the data completion method, and a data repair record is generated. A structured query statement is used to extract complete data from the four data tables, and a solution document containing a basic information area, a feature data area, and a scenario parameter area is generated, and the document version number and generation time are recorded. A retrieval mapping table is created for the generated solution document, which contains five fields: solution identification code, document version number, timestamp, data source table identification, and index type identification. The mapping table is stored in the database index area. A distributed storage mechanism is used to back up three copies of the solution document, and the copies are stored in storage nodes at different physical locations. The storage location identification and synchronization timestamp of each copy are recorded. The data management of retired power batteries involves the establishment and maintenance of multiple association tables. Taking a certain model of ternary lithium battery as an example, its basic information includes model identification LIR18650-30, capacity value 3000 mAh, and internal resistance value 20 milliohms. The original string "LIR18650-30_3000_20" is generated by information splicing, and the fixed-length identification code "7a8b9c0d" is generated by the SHA-256 hash algorithm, and the timestamp "2024-01-08-14:30:25" is recorded at the same time. The database design adopts a relational structure. The parameter information table contains basic parameter fields such as model, capacity, and internal resistance. The impedance spectrum data table stores frequency points, real and imaginary impedance values. The level information table records the ladder level and evaluation time. The scenario information table contains scenario parameters such as temperature range and rate limit. The scheme identification code is used as the primary key to establish a one-to-one association between tables. A timestamp index is established for the scheme time, a B-tree index is established for the battery model, and a hash index is established for the scenario type. The data integrity check found that there was an anomaly in the impedance spectrum data table, and the impedance value at the frequency point of 1000 Hz was missing. The data quality report records the abnormal location "Impedance spectrum data table-frequency 1000 Hz", the abnormal type "missing value", and the severity "medium". The linear interpolation method is used to fill in the missing data based on the impedance values ​​of the adjacent frequency points 999 Hz and 1001 Hz. The data repair record shows that the completed value is 0.82 milliohms for the imaginary part and 1.25 milliohms for the real part.The solution document adopts a partitioned storage structure. The basic information area records the battery model, capacity, and internal resistance. The characteristic data area stores the impedance spectrum curve parameters. The scenario parameter area contains the working condition requirements of the utilization scenario. The document version number adopts a three-segment format "1.0.0", where the first digit indicates the main version number, the second digit indicates the function update version number, and the last digit indicates the revision version number. The retrieval mapping table records the solution identification code "7a8b9c0d", version number "1.0.0", timestamp "20240108143025", data table identifier "EIS_DATA", and index type "TIME_INDEX". Data backup adopts a distributed storage mechanism to create document copies in three different storage nodes. The master node is located in the local storage server and is identified as "NODE_001". The two backup nodes are located in remote computer rooms and are identified as "NODE_002" and "NODE_003". Each copy records a synchronization timestamp to track data consistency. When the data of the master node changes, the data content of the backup node is updated through the asynchronous replication mechanism.

[0037] S107. In the actual step-by-step utilization process, according to the formed step-by-step utilization plan for retired batteries, collect the real-time working data of retired power batteries in corresponding utilization scenarios, including voltage, current, and temperature, and determine whether the battery is in an abnormal state through a preset abnormality detection model.

[0038] The acquisition unit in the battery management module is used to obtain the working parameters of the retired battery, and the working parameters include positive and negative electrode voltage values, charge and discharge current values, surface temperature values, and ambient temperature values; the battery state of charge value is obtained by the current integration method according to the charge and discharge current value; the Butterworth low-pass filter is used to obtain filtered data according to the working parameters, and the filtered data is processed by the Kalman filter to obtain a smoothed data sequence; the sliding window method is used to calculate the three indicators of voltage change rate, temperature change rate, and current fluctuation rate for the smoothed data sequence, and the thermal runaway risk, overcharge and over-discharge risk, and temperature abnormality risk are obtained by the weighted summation method according to the three indicators and the battery state of charge value; the thermal runaway risk, overcharge and over-discharge risk, and temperature abnormality risk are input into a pre-trained long short-term memory network to obtain a risk prediction value. If the risk prediction value exceeds a preset threshold, the isolation forest algorithm is used to analyze the abnormal parameter sequence to obtain abnormal alarm data.

[0039] Specifically, the acquisition unit in the battery management module is used to collect the working parameters of the retired battery according to a fixed sampling period, including the positive and negative voltage values, the charge and discharge current values, the surface temperature values, and the ambient temperature values. The battery state of charge value is calculated by the current integration method, and the Butterworth low-pass filter is used to reduce the noise of the collected data. The validity of the collected data after filtering is verified, the detection interval is set according to the battery specification parameters, the invalid data beyond the interval is eliminated, the parameter change rate of adjacent sampling points is calculated, and the Kalman filter is used to smooth the data sequence. The sliding window method is used to calculate the three indicators of voltage change rate, temperature change rate, and current fluctuation rate. Combined with the battery state of charge value, the three risk indicators of thermal runaway risk, overcharge and over-discharge risk, and temperature abnormality risk are calculated by the weighted summation method. The risk indicators are arranged according to the time series, and the feature sequence is constructed and input into the pre-trained long short-term memory network. The network includes an input layer, a hidden layer, and an output layer. The output layer nodes correspond to risk prediction values. The risk indicator threshold is determined based on the statistical results of historical abnormal data. If the risk prediction value exceeds the threshold, anomaly detection is triggered, and the exponential weighted average method is used to perform time series analysis on the parameter data before the abnormality occurs. The isolation forest algorithm is used to analyze the abnormal parameter sequence, calculate the abnormal score of the parameter point, compare the abnormal score with the preset threshold, locate the key abnormal point, and generate alarm data containing abnormal type, abnormal parameter, and abnormal time. According to the abnormal type in the alarm data, the corresponding parameter backtracking time interval is selected, the original parameter data in the time interval is extracted, and the multi-dimensional parameter data is feature extracted using the data dimension reduction method to generate abnormal traceability data. The operating status of retired power batteries needs to be monitored in real time during the cascade utilization process. Taking a certain model of ternary lithium battery as an example, the sampling period is set to 100 milliseconds, and parameters such as voltage, current, and temperature are collected. There is noise interference in the original collected data. For example, the voltage value fluctuates around 3.8 volts, and the fluctuation amplitude reaches 50 millivolts. A 4th-order Butterworth low-pass filter is used for noise reduction, and the cutoff frequency is set to 5 Hz. After filtering, the voltage fluctuation is reduced to less than 5 millivolts. The data validity verification is based on the detection interval set by the battery specification parameters. The voltage valid range is 2.5 to 4.2 volts, the current limit is 2 times the rated value, and the temperature range is -20 to 60 degrees Celsius. Data beyond the range is eliminated, such as the collected temperature value of 85 degrees Celsius, which is obviously beyond the normal range. The Kalman filter is used to smooth the data sequence, the prediction error covariance is set to 0.1, and the measurement error covariance is 0.2. After processing, a continuous and smooth parameter curve is obtained. The risk indicator calculation uses a sliding window of 60 seconds in length, and the voltage change rate, temperature change rate, and current fluctuation rate are calculated within the window. When the temperature rises by 8 degrees Celsius within 10 minutes, the voltage drops by 0.2 volts, and the current fluctuation exceeds 0.5 times the rated value, it indicates that there may be a risk of thermal runaway. Similarly, when the voltage exceeds 4.15 volts or is lower than 2.8 volts, it indicates the risk of overcharge and over-discharge. Temperature anomaly judgment is based on temperature mean and variance.The long short-term memory network uses a hidden layer containing 64 neurons, and the input features are the risk indicator sequence of the last 30 minutes. The network training data comes from historical abnormal cases, including typical failure modes such as overcharging, overdischarging, and temperature runaway. When the predicted risk value exceeds 0.8, the abnormal detection is triggered, and a 30-minute time window is used for parameter retrospective analysis. The isolation forest algorithm locates the abnormal parameters, sets the sampling size to 256, the tree depth to 8, and calculates the abnormal score for the parameter sequence. Parameter points with a score exceeding 0.6 are marked as abnormal points, and the specific time and parameter value of the abnormality are recorded. For example, in a certain test, it was found that the battery temperature rapidly rose by 10 degrees Celsius within 5 minutes, the voltage dropped by 0.3 volts at the same time, and the current fluctuated violently. The system determined it as a thermal runaway risk and generated an alarm message containing the abnormal type, time, and parameters. Retrospective analysis extracts parameter data 60 minutes before the abnormality occurs, and uses the principal component analysis method to reduce the dimension of multidimensional data such as voltage, current, and temperature, retaining feature components with a contribution rate of more than 95%. By restoring the data, it was found that 30 minutes before the abnormality occurred, the battery had shown signs of slowly rising temperature and increased internal resistance.

[0040] S108. If an abnormal state of the battery is detected, the early warning mechanism is triggered, and corresponding safety protection measures are taken according to the preset processing strategy, including cutting off the power supply and starting the cooling system.

[0041] According to the abnormality type, abnormal parameter value and abnormality degree recorded in the battery abnormality alarm data, the abnormality level is determined by comparing with the preset abnormality level classification table, and the corresponding protective measure instruction sequence is read from the processing strategy database; the main controller is used to generate a control instruction package in a standard format according to the protective measure instruction sequence, and the control instruction package is converted into an execution unit control signal; temperature control and overcharge and over-discharge protection measures are implemented through the execution unit, and the state feedback acquisition module is used to obtain the execution unit action state data stream, and the action state data stream includes relay disconnection feedback and refrigeration system working status; according to the action state data stream, a Bayesian network node is constructed to calculate the probability of successful execution of the protective measure, and a lossless compression algorithm is used to generate an abnormal processing report and store it in the archive database.

[0042] Specifically, according to the abnormal type, abnormal parameter value, and abnormal degree recorded in the battery abnormal alarm data, the abnormal level is determined by comparing with the preset abnormal level classification table, and the corresponding protective measure instruction sequence is read from the processing strategy database to generate a processing strategy list containing the instruction priority. A master-slave control structure is used to distribute control instructions to each execution unit. The main controller generates a control instruction package in a standard format according to the processing strategy list, and the control instruction is converted into a control signal of the execution unit through the protocol conversion module. For temperature abnormality protection measures, a refrigeration control parameter sequence is generated according to the refrigeration power gear, cooling fan speed, and refrigeration cycle duration, and the parameter sequence is converted into a pulse width modulation signal through a digital signal processor. For overcharge and over-discharge protection measures, a relay disconnection signal and a balancing circuit conduction signal are generated, and an optoelectronic isolation module is used to electrically isolate the control signal, and an execution signal is output through a power driver. A state feedback acquisition module is used to record the action status of the execution unit, including relay disconnection feedback, refrigeration system working status, cooling fan speed value, and actuator action time, and an execution status data stream is generated. According to the execution status data flow, Bayesian network nodes are constructed, including execution action completion nodes, execution timing compliance nodes, and execution effect compliance nodes, and the probability of successful execution of protective measures is calculated through probabilistic reasoning. The execution data of the protective measures are recorded, including abnormal trigger conditions, control instruction content, execution action parameters, and execution effect data. The abnormal processing report is generated by a lossless compression algorithm and stored in the archive database. The execution of abnormal protection measures for retired power batteries requires strict control timing and reliable execution verification. Taking the ternary lithium battery pack of a certain energy storage power station as an example, when it is detected that the battery temperature rises by 12 degrees Celsius within 5 minutes, and the voltage drops rapidly by 0.3 volts, the system determines that it is a thermal runaway risk, and the abnormal level is level one. At this time, the protection instruction sequence is read from the processing strategy database, including three instructions: disconnecting the charging and discharging circuit, starting forced cooling, and switching the backup power supply, and the instruction priorities are 1, 2, and 3 respectively. The main controller adopts a multi-task parallel processing architecture to convert the control instruction into a control frame in a standard format. The control frame contains the instruction type, target address, action parameters, and checksum. The control frame is converted into the dedicated protocol of each execution unit through the protocol conversion module, such as the Modbus protocol for the refrigeration controller and the CAN protocol for the relay controller. In view of the risk of thermal runaway, the control parameters of the refrigeration system include compressor power, coolant flow, and cooling fan speed. The compressor power is set to 90% of the rated value, corresponding to an output frequency of 45 Hz, the coolant flow is set to 20 liters per minute, and the cooling fan speed is set to 3000 rpm. These parameters are converted into a pulse width modulation signal with a duty cycle of 80% and a frequency of 20 kHz by a digital signal processor. The power cut-off adopts a dual-channel relay series structure, and the drive signals of the two relays are electrically isolated by 4000 volts through a photoelectric coupler.The relay coil drive voltage is 12 volts, the contact current is 50 amperes, and the gate drive signal output by the power driver controls the relay action. To prevent arcing when disconnected, a RC absorption circuit is connected in parallel to the relay contact. The execution status feedback acquisition module records the action status of each execution unit. The relay status is fed back through the auxiliary contact, the refrigeration system is fed back through the temperature, pressure, and flow sensors, and the cooling fan is fed back through the speed sensor. The Bayesian network uses a three-layer structure to evaluate the execution effect. The bottom layer nodes correspond to the completion of specific execution actions, the middle layer nodes indicate the degree of compliance with the execution sequence, and the top layer nodes reflect the overall protection effect. When all thermal runaway protection measures are executed, the battery temperature drops by 8 degrees Celsius within 10 minutes, the voltage tends to stabilize, and the system calculates that the probability of successful execution of the protection measures is 0.95. The exception handling report records the trigger conditions, control process, execution effect and other data. The report is compressed to 40% of the original size using Huffman coding and stored in the archive. These data provide an important basis for subsequent optimization of protection strategies.

[0043] Based on the above embodiments of the present invention, relevant personnel can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for predicting the remaining life of retired power cells, characterized in that: The method comprises: Obtain the basic parameter information of the battery cell to be tested in the retired power battery, including the battery cell model, capacity, and internal resistance, and obtain the standard impedance spectrum data of the battery cell of this model from the preset battery cell parameter database according to the battery cell model; For the cells to be tested in the retired power batteries, impedance tests at different frequencies are performed within a preset test frequency range to obtain impedance response data at different frequencies and obtain impedance spectrum data; Preprocess the acquired impedance spectrum data, including removing abnormal points, smoothing data, extracting key characteristic parameters of the impedance spectrum curve, including arc radius, center coordinates, slope, and constructing impedance spectrum feature vectors; The impedance spectrum feature vector is input into the pre-built life prediction model, and the predicted remaining life data of the battery cell to be tested is output. Combined with the preset step utilization division rule, the step level to which the battery cell to be tested belongs is obtained; Obtain the usage scenario conditions corresponding to different levels, including temperature range and charge / discharge rate, and match the target level usage scenario in combination with the level to which the current battery cell to be tested belongs; The target step-by-step utilization scenario is associated with the retired battery cell model, capacity, internal resistance and impedance spectrum data information to form a step-by-step utilization plan for retired batteries; In the actual step-by-step utilization process, according to the formed step-by-step utilization plan for retired batteries, the real-time working data of retired power batteries in the corresponding utilization scenarios, including voltage, current, and temperature, are collected, and the preset abnormality detection model is used to determine whether the battery is in an abnormal state; If an abnormal battery condition is detected, the early warning mechanism will be triggered and corresponding safety protection measures will be taken according to the preset processing strategy, including cutting off the power supply and starting the cooling system.

2. The method according to claim 1, characterized in that The basic parameter information of the battery cell to be tested in the retired power battery is obtained, including the battery cell model, capacity, and internal resistance, and the standard impedance spectrum data of the battery cell of this model is obtained from a preset battery cell parameter database according to the battery cell model, including: Acquire design specification parameters from a preset database according to the retired power cell identification, wherein the design specification parameters include a nominal capacity value, a rated voltage value, and a standard internal resistance value; For retired power cells, constant current and constant voltage are used to charge them to the rated voltage value, and stable cells are obtained by standing them for a preset time. The electrochemical impedance spectrometer is used to perform frequency sweep measurement on the stable state battery cell to obtain raw impedance data, and a real and imaginary impedance numerical sequence is generated by fast Fourier transform; The impedance value sequence is verified for data validity, abnormal data points are removed, and missing data are supplemented using cubic spline interpolation to generate a continuous impedance curve.

3. The method according to claim 1, characterized in that The method includes performing impedance tests at different frequencies within a preset test frequency range on the cells to be tested in the retired power battery, obtaining impedance response data at different frequencies, and obtaining impedance spectrum data, including: Apply constant current and constant voltage to charge the retired power cells to the charging cut-off voltage value, and use a constant current power supply to apply rated current to the charged cells to discharge to the discharge cut-off voltage value; Monitor the open circuit voltage of the discharged battery cell, and determine that the battery cell is in a stable state when the open circuit voltage fluctuation value is less than a preset voltage fluctuation threshold; Applying a sinusoidal AC excitation signal to a battery cell in a stable state by using an electrochemical impedance spectrometer, and performing Fourier spectrum conversion on the excitation signal response to obtain an impedance value sequence; The impedance data curve is obtained by performing point filling processing on the impedance numerical sequence using a cubic spline interpolation method, and the positions of semicircular arc feature points are identified according to the impedance data curve to obtain an impedance spectrum feature data set.

4. The method according to claim 1, characterized in that The acquired impedance spectrum data is preprocessed, including removing abnormal points, smoothing data, extracting key characteristic parameters of the impedance spectrum curve, including arc radius, center coordinates, slope, and constructing an impedance spectrum feature vector, including: The local variance value of the impedance spectrum data sequence is calculated using a sliding window. If the variance value exceeds the mean threshold of all variances in the frequency interval, the marked abnormal data point is removed from the impedance spectrum data sequence. The interval value between adjacent frequency points is calculated based on the impedance spectrum data sequence after removing abnormal points. If the interval exceeds the preset frequency step, the linear interpolation method is used to obtain the supplementary frequency point data; For the impedance spectrum sequence after the supplemented frequency point data, high-frequency noise components are removed by using a Butterworth low-pass filter and four-layer discrete wavelet decomposition, and a continuous and smooth impedance spectrum curve is obtained by using a cubic spline function; According to the continuous and smooth impedance spectrum curve, arc fitting and linear fitting are performed on the complex plane, the coordinate value of the center of the circle and the radius value are obtained by iterative least square method, the real part value and imaginary part value of the low-frequency slope are obtained by complex number operation, and the characteristic vector is normalized.

5. The method according to claim 1, characterized in that The impedance spectrum feature vector is input into a pre-built life prediction model, and the predicted remaining life data of the battery cell to be tested is output, and the ladder level to which the battery cell to be tested belongs is obtained in combination with a preset ladder utilization division rule, including: The mean and standard deviation of each characteristic component are calculated according to the characteristic vector of the impedance spectrum, and the standardized characteristic vector is obtained by using the Gaussian distribution standardization method; Performing singular value decomposition on the standardized feature vector, sorting by singular value size and retaining feature components whose cumulative contribution rate exceeds a contribution rate threshold, to obtain a reduced-dimensional feature vector; The dimension-reduced feature vector is input into a preset deep neural network, and the output value of each layer node is nonlinearly transformed to obtain the predicted value of the remaining life of the battery cell; The fuzzy clustering method is used to calculate the membership degree of each level for the predicted value of the remaining life of the battery cell. The grading result is verified by calculating the Mahalanobis distance between the characteristic vector of the battery cell and the center of each level of samples to obtain the final grading data of the battery cell.

6. The method according to claim 1, characterized in that The obtaining of the utilization scenario conditions corresponding to different steps, the utilization scenario conditions including the temperature range and the charge and discharge rate, and matching the target step utilization scenario in combination with the step level to which the current battery cell to be tested belongs, includes: Reading a step-level utilization scenario table from a preset scenario parameter database, wherein the step-level utilization scenario table includes a temperature range limit, a charge and discharge rate limit, a voltage range limit, a cycle depth limit, and a working time limit; According to the step levels, a judgment matrix is ​​constructed using the limit parameters in the scenario table, and weight coefficients of the limit parameters are calculated by a hierarchical analysis method; Extracting the operating parameters of the energy storage scenario, the power scenario and the backup power scenario according to the weight coefficient, and obtaining the scenario adaptability score by using a data statistical method; The membership value of the battery cell to be tested to each scenario is calculated according to the scenario adaptability score, and the Monte Carlo method is used to verify whether the performance parameters of the battery cell to be tested meet the scenario requirements.

7. The method according to claim 1, characterized in that The step-by-step utilization scenario is associated with the retired battery cell model, capacity, internal resistance and impedance spectrum data information to form a retired battery step-by-step utilization plan, including: The model identification, capacity value, internal resistance value and impedance spectrum data are concatenated to generate the original identification string, and the solution identification code is obtained through the secure hash algorithm; Establish a parameter information table, an impedance spectrum data table, a level information table and a scenario information table according to the scheme identification code, and use the scheme identification code as a primary key to establish an association relationship between the data tables; Using a null value detection method to identify missing data items in a data table, obtaining abnormal data markers through a data type verification method, and generating a data quality report for the abnormal data markers; A structured query statement is used to extract complete data from a data table, and a retired battery step-by-step utilization plan document including a basic information area, a feature data area, and a scenario parameter area is generated based on the complete data.

8. The method according to claim 1, characterized in that In the actual step-by-step utilization process, according to the formed step-by-step utilization plan for retired batteries, real-time working data of retired power batteries in corresponding utilization scenarios are collected, including voltage, current, and temperature, and a preset abnormality detection model is used to determine whether the battery is in an abnormal state, including: The acquisition unit in the battery management module is used to obtain the working parameters of the retired battery, wherein the working parameters include positive and negative electrode voltage values, charge and discharge current values, surface temperature values ​​and ambient temperature values; Obtaining a battery state of charge value by a current integration method according to the charge and discharge current values; A Butterworth low-pass filter is used to obtain filtered data according to the working parameters, and the filtered data is processed by a Kalman filter to obtain a smoothed data sequence; The sliding window method is used to calculate the voltage change rate, temperature change rate and current fluctuation rate of the smoothed data sequence, and the thermal runaway risk, overcharge and overdischarge risk and temperature abnormality risk are obtained by weighted summation method according to the three indicators and the battery state of charge value; The thermal runaway risk, overcharge and over-discharge risk, and temperature abnormality risk are input into a pre-trained long short-term memory network to obtain a risk prediction value. If the risk prediction value exceeds a preset threshold, the isolation forest algorithm is used to analyze the abnormal parameter sequence to obtain abnormal alarm data.

9. The method according to claim 1, characterized in that: If the battery is detected to be in an abnormal state, the early warning mechanism is triggered, and corresponding safety protection measures are taken according to the preset processing strategy, including cutting off the power supply and starting the cooling system, including: According to the abnormal type, abnormal parameter value and abnormal degree recorded in the battery abnormal alarm data, the abnormal level is determined by comparing with the preset abnormal level classification table, and the corresponding protective measure instruction sequence is read from the processing strategy database; Using a main controller to generate a control instruction packet in a standard format according to the protective measure instruction sequence, and converting the control instruction packet into an execution unit control signal; Implement temperature control and overcharge and overdischarge protection measures through the execution unit, and use the state feedback acquisition module to obtain the execution unit action state data stream, which includes relay disconnection feedback and refrigeration system working status; A Bayesian network node is constructed according to the action state data stream to calculate the probability of successful execution of protective measures, and an exception handling report is generated using a lossless compression algorithm and stored in an archive database.

Images