A health assessment method and system for retired batteries

Abstract

The present invention discloses a health assessment method and system for retired batteries, which relates to the field of battery status assessment technology. The method includes identifying the retired battery identifier and automatically generating an assessment task number, setting the battery status to "pending detection"; setting the battery charging mode, collecting battery status data in real time, and extracting battery features to construct a feature vector G; inputting the feature vector G into a parallel attention neural network for static SOH estimation to obtain a preliminary SOH estimate, using a joint extended Kalman filter to perform SOH correction, and obtaining a current equivalent internal resistance estimate. The present invention can achieve efficient and accurate assessment of the health status of retired batteries, significantly improves the comprehensiveness of feature extraction and deep modeling capabilities, enhances the reliability and dynamic adaptability of SOH estimation results, effectively reduces error fluctuations, and improves the assessment system's generalization ability for complex retirement states.

Description

Technical Field

[0001] The present invention relates to the technical field of battery status assessment, and in particular to a health assessment method and system for retired batteries. Background Art

[0002] With the rapid development of the new energy vehicle industry and renewable energy systems, lithium-ion batteries, as core energy storage components, are widely used in multiple scenarios, including power batteries, grid energy storage, and power tools. However, after a certain number of charge and discharge cycles, batteries will gradually reach their performance degradation threshold and be retired. According to existing industry data, a large market for retired batteries will be formed in the next few years. To improve resource utilization, reduce environmental pressure, and promote cascade utilization applications, there is an urgent need to conduct scientific and effective health assessments of retired batteries. Existing assessment methods mostly rely on static capacity testing, internal resistance measurement, or full-life cycle experiments. These methods not only have long test cycles and high operating costs, but are also difficult to adapt to the complex sources and highly discrete performance of retired batteries. In recent years, with the integration of big data and artificial intelligence technologies in battery management systems, data-driven SOH estimation and life prediction methods have developed rapidly. However, in retired battery health assessment scenarios, they still face prominent problems such as poor model generalization, redundant feature dimensions, insufficient estimation accuracy, and lack of multi-source decision fusion.

[0003] Existing technologies suffer from limitations in several key areas. First, in terms of feature extraction, most methods rely solely on simple statistical features, such as constant current duration and termination voltage, failing to uncover the deeper health characteristics hidden in time-series data such as voltage and current. In particular, they lack comprehensive utilization of wavelet frequency domain, differential signals, and dynamic trend indicators. Second, in terms of SOH estimation methods, existing technologies often use single models (such as neural networks or Kalman filters) to process static data. They lack mechanisms to integrate the strengths of multiple models for joint modeling and adaptive correction, making it difficult to balance prediction accuracy with dynamic tracking capabilities. Third, feature selection often relies on manual experience or simple filtering, lacking intelligent, optimization-driven feature selection strategies. This can easily miss key features or introduce redundant information, impacting model performance. Finally, in terms of assessment output, results are often presented as a single metric, lacking systematic reporting and platform integration capabilities, making them incapable of meeting the industry's needs for full battery lifecycle management. Consequently, existing retired battery health assessment technologies suffer from weak modeling capabilities, low prediction accuracy, poor feature utilization, and insufficient platform integration, making them incapable of supporting future large-scale engineering applications for the cascade utilization of retired batteries. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a health assessment method and system for retired batteries, which solves the problems of weak modeling capabilities, low prediction accuracy, poor feature utilization and insufficient platform integration commonly found in existing retired battery health assessment technologies, making it difficult to support the engineering application of large-scale retired battery recycling in the future.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a health assessment method for retired batteries, comprising:

[0008] Identify the retired battery logo and automatically generate an assessment task number, setting the battery status to "pending inspection";

[0009] Set the battery charging mode, collect battery status data in real time and extract battery features to construct feature vector G;

[0010] The battery status data includes voltage, current and time data;

[0011] The feature vector G is input into the parallel attention neural network for static SOH estimation to obtain a preliminary SOH estimate. The joint extended Kalman filter is used to correct the SOH and obtain the current equivalent internal resistance estimate.

[0012] Estimate the maximum available power of the battery based on the current equivalent internal resistance estimate and real-time voltage , based on SOH and Classify battery health levels and recommend application scenarios;

[0013] All data are packaged into assessment reports and archived in the cloud.

[0014] As a preferred embodiment of the health assessment method for retired batteries of the present invention, wherein: setting the battery charging mode, collecting battery status data in real time and extracting battery features to construct a feature vector G refers to collecting battery voltage, current and time data and recording them as an original data set;

[0015] After preprocessing the data in the original data set, the primary features of the data are extracted, including the charging time in the constant current stage , Charging time in constant voltage stage , voltage mid-segment rise rate , incremental capacity main peak , Corresponding voltage position and the peak time point of the differential voltage curve ;

[0016] Integrate the extracted features to form a first-level feature set ;

[0017] The voltage and current in the data set are decomposed into three layers of wavelet using Daubechies4 wavelet basis function to obtain six groups of sub-signals. For each sub-signal, features are extracted and integrated to generate a wavelet feature set. ;

[0018] The feature set and Merge and construct the total feature set F;

[0019] Perform a full discharge on each battery to obtain the true capacity label a. Use the Pearson correlation evaluation method to calculate the correlation between each dimension feature and the true capacity label. Retain the features with correlation greater than the preset threshold and integrate them into the filtered feature set F'. Construct a binary selection variable for each feature in F' to construct a binary vector of length k ,in: Indicates whether the kth feature in the full feature set is selected, and each B corresponds to a feature subset combination;

[0020] Use the sparrow search algorithm to search for the optimal feature subset, and divide the individual roles into the first The role of the individual is the discoverer, and the remaining individuals are followers;

[0021] Based on the feature data, the XGBoost regressor is used to obtain the predicted capacity, and the average error between the predicted capacity and the actual capacity is calculated as the fitness value. The individual with the lowest fitness is set as the global optimal solution. ;

[0022] For the discoverer, an exponential decay perturbation update strategy is implemented;

[0023] Perform global guidance adjustment on the remaining individuals with reference to the current global optimal solution;

[0024] The individual with the highest current fitness is marked as a vigilant, and mutation perturbation is performed to escape the local optimum;

[0025] The continuous coding vectors of all individuals are probability mapped through the Sigmoid function, and then thresholded and binarized to obtain the updated binary coding individuals. After reaching the maximum number of iterations W, the final global optimal coding solution A is output. According to the optimal coding A, the selected feature dimensions are extracted to form the optimal feature subset, which is normalized and then spliced into the feature vector G.

[0026] As a preferred embodiment of the health assessment method for retired batteries according to the present invention, the step of inputting the feature vector G into the parallel attention neural network for static SOH estimation to obtain a preliminary SOH estimate comprises:

[0027] Divide the feature vector G into two sub-vectors according to the feature source, including the time series feature vector and morphological feature vectors , respectively used for different model channel processing;

[0028] from The main variables reflecting the evolution of battery charge characteristics are extracted and recorded as signal sequence x(t). The empirical mode decomposition algorithm is used to decompose the signal sequence x(t) into several intrinsic mode functions and residuals.

[0029] Select the first two order IMF components and Used to reconstruct the main sequence ;

[0030] Combine the features of all time steps into the input tensor , where t is the time step and d is the number of input feature dimensions at each time step;

[0031] Use the LSTM model to build the time channel, build the first layer of LSTM-1, initialize the input dimension, number of hidden units, activation function, and input Use a linear layer to increase the dimension to the same size as the residual dimension and normalize the output vector at each time step;

[0032] The normalized output vector is fed into the second layer LSTM-2 network, and the input Use a linear layer to increase the dimension to the same dimension to match the residual dimension, normalize the output vector of each time step, perform a time average pooling operation on the output vector, send the pooled vector to the fully connected network, and output it by the output layer;

[0033] Collect historical battery data as a training set and input it into the LSTM model for iterative training. Define the mean square error as the training target loss function and use the Adam optimizer to iteratively optimize the model parameters. If the error does not decrease after P consecutive rounds of verification, stop the iteration and obtain the optimal model parameters as the initial parameter set E.

[0034] Expand E into a one-dimensional vector as the initial position of the particle and initialize the APSO group, set the group size, and initialize a velocity vector for each particle at the same time;

[0035] Start iterative optimization, set the maximum number of rounds, perform speed and position updates for each particle, calculate the fitness value corresponding to each particle, and update the individual optimal solution and the global optimal solution;

[0036] After completing the global search iteration of APSO, the current global optimal particle position is directly extracted as the final optimization solution and restored to the LSTM model parameter structure. The optimized parameters are loaded into the original LSTM model to obtain the optimized model;

[0037] Will Input the optimized LSTM model to obtain the preliminary SOH prediction value;

[0038] Use Attention to construct the morphological feature channel, set the multi-head attention parameters, including the number of attention heads and the output dimension of each head, and construct three sets of mapping matrices for each attention head. Perform linear transformation to obtain query Q, key K, value V tensors, perform self-attention calculation on each attention head, and concatenate the output vectors of all heads along the dimension to form the final attention modeling output, which represents the health status representation vector modeled by the current morphological feature channel;

[0039] Receive the output of the time channel and morphological feature channel and splice them to form a fusion vector , construct the MLP network structure, including input layer, hidden layer and output layer, Input into MLP to get a preliminary SOH estimate .

[0040] As a preferred embodiment of the health assessment method for retired batteries according to the present invention, the method of using a joint extended Kalman filter to perform SOH correction and obtain a current equivalent internal resistance estimate includes:

[0041] Collect battery factory rated capacity , calculate the initial capacity estimate based on the SOH estimate and rated capacity;

[0042] Construct and initialize the filter state vector, which includes the current state of charge, RC voltage term, current equivalent internal resistance, and current battery capacity;

[0043] At the same time, the covariance matrix, state noise and observation noise are initialized based on historical data;

[0044] Extract dynamic observation data sequences from the standard charging test process and use the battery RC equivalent model to predict state variables, including the state of charge and RC voltage terms.

[0045] Calculate the current predicted voltage based on the state prediction value ;

[0046] Calculate the error between the actual voltage and the predicted voltage;

[0047] Update the state vector and state covariance using the standard EKF Kalman gain formula;

[0048] After all sampling points are iterated, the final state vector is estimated, including the current state of charge estimate, RC voltage term estimate, current equivalent internal resistance estimate, and current battery capacity estimate.

[0049] Extract the battery's current capacity estimate from the final state vector estimate and calculate the final health estimate using the battery's current capacity estimate and rated capacity. .

[0050] As a preferred solution of the health assessment method for retired batteries of the present invention, wherein: the maximum available power of the battery is estimated based on the current equivalent internal resistance estimate and the real-time voltage , based on SOH and Battery health level classification and application scenario recommendations include:

[0051] Extract the current equivalent internal resistance estimate from the final state vector estimate , and estimate the maximum available power of the battery based on the real-time voltage U ;

[0052] The final health valuation and the maximum available power of the battery are integrated to generate an input feature vector. A random forest is used as the classification model, and health level labels are defined as A, B, and C, corresponding to high health, medium health, and low health, respectively. The input feature vector is input into the random forest model to obtain the corresponding health level label, and the corresponding application scenario is automatically recommended based on the classification results.

[0053] As a preferred embodiment of the health assessment method for retired batteries described in the present invention, the identification of retired battery identification and automatic generation of an assessment task number, and setting the battery status to "pending detection" refer to the process of reading the RFID tag information attached to the battery by an RFID scanning device during the battery warehousing process and using it as the unique number of the battery. After obtaining the battery number, the assessment task number corresponding to the battery is synchronously generated, the assessment task number and the battery number are bound one by one, and after the information is bound, the initial status of the task is set to "pending detection".

[0054] As a preferred solution of the health assessment method for retired batteries described in the present invention, the encapsulation of all data into an assessment report and archiving and storing it in the cloud refers to integrating the basic information of the battery and the data generated by the analysis, generating a complete battery health assessment report, and uploading it to the cloud database platform.

[0055] In a second aspect, the present invention provides a health assessment system for retired batteries, comprising:

[0056] The task management module is used to identify the retired battery identification and automatically generate an assessment task number, setting the battery status to "pending inspection";

[0057] Feature extraction module, used to set the battery charging mode, collect battery status data in real time and extract battery features to construct feature vectors;

[0058] The estimation module is used to input the feature vector into the parallel attention neural network for static SOH estimation to obtain a preliminary SOH estimate;

[0059] A correction module is used to perform SOH correction using a joint extended Kalman filter and obtain the current equivalent internal resistance estimate;

[0060] The grading module estimates the maximum available power of the battery based on the current equivalent internal resistance and real-time voltage, and performs health grading and application scenario recommendations based on SOH.

[0061] The archiving module is used to package all data into assessment reports and archive them in the cloud.

[0062] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the health assessment method for retired batteries as described in the first aspect of the present invention is implemented.

[0063] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the health assessment method for retired batteries as described in the first aspect of the present invention.

[0064] The beneficial effects of the present invention are as follows: the present invention introduces a multi-scale feature extraction mechanism based on wavelet decomposition and correlation evaluation, combines sparrow search with XGBoost to realize optimal feature subset screening, and constructs a parallel neural network integrating LSTM and multi-head attention structure for SOH estimation, and then uses the joint extended Kalman filter to realize state correction and internal resistance valuation, which can realize efficient and accurate evaluation of the health status of retired batteries, significantly improve the comprehensiveness of feature extraction and deep modeling capabilities, enhance the reliability and dynamic adaptability of SOH estimation results, effectively reduce error fluctuations, and improve the generalization ability of the evaluation system for complex retirement states. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0066] Figure 1 Flowchart of the health assessment method for retired batteries in Example 1.

[0067] Figure 2 This is a structural diagram of the health assessment system for retired batteries in Example 1.

[0068] Figure 3 Schematic diagram of the feature optimization process in Example 1. DETAILED DESCRIPTION

[0069] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0070] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0072] Example 1, with reference to Figures 1 to 3 , which is the first embodiment of the present invention, provides a health assessment method for retired batteries, comprising the following steps:

[0073] S1. Identify the retired battery logo and automatically generate an assessment task number, and set the battery status to "pending inspection";

[0074] Specifically, it identifies the retired battery identification and automatically generates an assessment task number. Setting the battery status to "pending inspection" means that during the battery storage process, the RFID tag information attached to the battery is read by the RFID scanning device and used as the battery's unique number. If the reading fails (such as the absence of the tag or the tag is damaged), a unique number based on the current date, test station and task number is automatically generated to ensure that each battery has an independent identification;

[0075] After obtaining the battery number, the evaluation task number corresponding to the battery is generated synchronously. The task number consists of the current date and the evaluation serial number and is unique. The evaluation task number is bound to the battery number one by one. After the information is bound, the initial status of the task is set to "pending detection".

[0076] Through RFID or the system's automatic numbering mechanism, each battery is guaranteed to obtain a unique identification upon entering the evaluation system, avoiding data confusion caused by number duplication, loss, or manual entry errors. The automatic generation and binding of evaluation task numbers enables the battery to have a unified indexing system for the entire process from identification to detection. Since some retired batteries may have problems such as missing RFID tags, damaged tags, or information falling off, the traditional method of relying on manual registration or scanning code entry is inefficient and has a high error rate. The present invention designs a numbering system based on a combination of time and workstation number generation, which can ensure unique number allocation even in the absence of physical identification, thereby enhancing the system's adaptability and fault tolerance to abnormal battery conditions.

[0077] S2. Set the battery charging mode, collect battery status data in real time, and extract battery features to construct a feature vector G;

[0078] Specifically, the battery charging mode is set, battery status data is collected in real time, and battery features are extracted to construct a feature vector G. This means placing the battery under test in a temperature-controlled test chamber, setting the test environment temperature, and starting the charging process (for example, charging at a constant current of 0.5C until the voltage reaches 4.20V, maintaining the voltage at 4.20V, and continuing to charge until the current drops to 0.05A). During the entire charging process, the battery voltage, current, and time data are collected and recorded as the original data set;

[0079] After preprocessing the data in the original data set, the primary features of the data are extracted, including the charging time in the constant current stage , Charging time in constant voltage stage , The rate of rise of the voltage in the middle section (for example, the average rate at which the voltage rises from 3.85V to 4.00V) , incremental capacity main peak :

[0080]

[0081] Where, is the charge, is the voltage;

[0082] Corresponding voltage position and the peak time point of the differential voltage curve :

[0083] ;

[0084] Where, I is the current in the constant current stage, T is the time;

[0085] Integrate the extracted features to form a first-level feature set ;

[0086] The voltage and current in the data set are decomposed into three layers of wavelet using Daubechies4 (db4) wavelet basis function to obtain six groups of sub-signals. For each sub-signal, 16 categories of statistical features are extracted to obtain 128-dimensional DWT features. The extracted features are integrated to generate a wavelet feature set. ;

[0087] The feature set and Merge and construct the total feature set F;

[0088] Perform a full discharge on each battery to obtain the true capacity label a. Use the Pearson correlation evaluation method to calculate the correlation between each dimension feature and the true capacity label. Features with correlation greater than a preset threshold (set by statistical methods) are retained and integrated into the filtered feature set F'. For each feature in F', a binary selection variable is constructed. , indicating whether to select the final subset, constructing a binary vector of length o ,in: : Indicates that the oth feature is retained; Indicates whether the oth feature in the feature set is selected;

[0089] : Indicates that the oth feature is not retained;

[0090] o is the remaining characteristic dimension after the initial screening;

[0091] Each B corresponds to a feature subset combination;

[0092] Use the sparrow search algorithm to search for the optimal feature subset, initialize the population size to N, the maximum number of iterations to W, and divide the individual roles into the front The role of the individual is the discoverer, and the remaining individuals are followers. Each individual is initially encoded as a binary vector of length d, where p is the discoverer ratio coefficient, It represents the number of individuals calculated in proportion, rounded to an integer (usually rounded down);

[0093] Based on the feature data, the XGBoost regressor is used to obtain the predicted capacity, and the average error between the predicted capacity and the actual capacity is calculated as the fitness value. The individual with the lowest fitness is set as the global optimal solution. ;

[0094] For the finder, an exponential decay perturbation update strategy is implemented to improve the search accuracy:

[0095]

[0096] Where, is the continuous position vector of the i-th individual in the t-th round, is the updated position of the i-th individual, is the step size factor that controls the exponential decay speed, W is the maximum number of iterations, is the local disturbance intensity factor, is a standard normally distributed random number, and Parameter setting through experiments;

[0097] Perform global guidance adjustment on the remaining individuals with reference to the current global optimal solution:

[0098]

[0099] Where, is the current position of the j-th follower, is the continuous code of the best individual in the current population, is the guiding influence factor, set by the grid search method, is a uniformly distributed random number, is the updated position of the follower individual;

[0100] The individual with the highest current fitness is marked as a vigilant, and mutation perturbation is performed to jump out of the local optimum:

[0101]

[0102] Where, It is The position of the individual with the worst fitness, is the position of an individual randomly selected from the current population, is the jump perturbation amplitude factor, set by the differential optimization algorithm, sign() is the sign function, is the post-jump position of the individual sentinel;

[0103] The continuous coding vectors of all individuals are probability mapped through the Sigmoid function, and then thresholded and binarized to obtain the updated binary coding individuals. After reaching the maximum number of iterations W, the final global optimal coding solution A is output. According to the optimal coding A, the selected feature dimensions are extracted to form the optimal feature subset, which is normalized and then spliced into the feature vector G.

[0104] The first-level feature set in the feature vector G directly characterizes the electrochemical behavior and performance response of the battery during charging. The DWT feature extracts six sub-signals by performing a three-layer Daubechies4 wavelet decomposition on the voltage and current curve signals. Statistical parameters such as energy, entropy, skewness, and kurtosis are then extracted to further reveal the health signal characteristics at different frequency bands. This joint time-frequency feature representation effectively captures the nonlinear and non-stationary characteristic evolution of retired batteries. The SSA optimization objective function is the mean absolute error (MAE) between the predicted capacity and the actual capacity of the XGBoost regressor. This allows the feature subset search to not only optimize dimensionality but also be deeply coupled with specific task performance, thereby enhancing model adaptability. Ultimately, the optimal feature subset is normalized to form a numerical feature vector G, which is compact, highly discriminative, and model-friendly, providing accurate input for the subsequent static SOH estimation model and dynamic state filter.

[0105] This technical solution has the following beneficial effects: First, through the fusion of wavelet decomposition and statistical features, feature extraction is expanded from the traditional single time domain to the time-frequency domain, thereby enhancing the ability to capture implicit health degradation information; second, the use of correlation screening and intelligent optimization algorithms to collaboratively perform feature selection significantly improves the effectiveness and stability of the selected feature set and reduces the risk of model overfitting; third, the constructed feature vector G is structurally convenient for input into deep model processing, supports the subsequent temporal modeling and morphological recognition of neural networks, and improves the accuracy and reliability of the entire health assessment process; fourth, the overall process has a high degree of automation, which is suitable for engineering deployment and batch evaluation needs, and is particularly suitable for the rapid sorting of retired batteries before cascade utilization.

[0106] S3. Input the feature vector G into the parallel attention neural network for static SOH estimation to obtain a preliminary SOH estimate, use the joint extended Kalman filter to perform SOH correction, and obtain the current internal resistance;

[0107] Specifically, the feature vector G is input into the parallel attention neural network for static SOH estimation, and the preliminary SOH estimation values are obtained including:

[0108] The feature vector G is divided into two sub-vectors according to the feature source, including the time series feature vector ( ) and morphological feature vectors (Statistical features such as energy, mean, kurtosis, skewness, and entropy extracted from six groups of sub-signals under wavelet decomposition (such as db4)) are used for processing different model channels, including time channel and morphology channel;

[0109] from Extract the main variables reflecting the evolution of battery charge characteristics (such as voltage sequence, voltage rise segment, dQ / dV interval), record them as signal sequence x(t), and use the empirical mode decomposition algorithm to decompose the signal sequence x(t) into several intrinsic mode functions and residuals:

[0110]

[0111] Where, is the original signal sequence, is the jth eigenmode function, It is the residual signal left after extracting the nth IMF, representing the trend component, n is the total number of IMFs, and t is the time step;

[0112] Select the first two order IMF components and Used to reconstruct the main sequence :

[0113]

[0114] Will As a new time series input, uniformly replace the original mid-sequence features;

[0115] Combine the features of all time steps into the input tensor , where t is the time step and d is the number of input feature dimensions at each time step;

[0116] Use the LSTM model to build the time channel, build the first layer of LSTM-1, initialize the input dimension, number of hidden units, activation function, and input Use a linear layer to increase the dimension to the same size as the residual dimension and normalize the output vector at each time step;

[0117] The normalized output vector is fed into the second layer of LSTM-2 network to initialize the input dimension, number of hidden units and output shape, and the residual mechanism is repeated (the input Use a linear layer to increase the dimension to the same size as the residual dimension, normalize the output vector of each time step), perform a time average pooling operation on the output vector, send the pooled vector to the fully connected network, and output it by the output layer;

[0118] Collect historical battery data as a training set and input it into the LSTM model for iterative training. Define the mean square error as the training target loss function and use the Adam optimizer to iteratively optimize the model parameters. If the error does not decrease after P consecutive rounds of verification, stop the iteration and obtain the optimal model parameters as the initial parameter set E. The minimum verification RMSE during the training process is retained as a reference for fitness evaluation.

[0119] Expand E into a one-dimensional vector as the initial position of the particle and initialize the APSO group, set the group size, and initialize a velocity vector for each particle at the same time;

[0120] Start iterative optimization, set the maximum number of rounds, perform speed and position updates for each particle, calculate the fitness value corresponding to each particle, and update the individual optimal solution and the global optimal solution;

[0121] After completing the global search iteration of APSO, the current global optimal particle position is directly extracted as the final optimization solution and restored to the LSTM model parameter structure. The optimized parameters are loaded into the original LSTM model to obtain the optimized model;

[0122] Will Input the optimized LSTM model to obtain the preliminary SOH prediction value;

[0123] Use Attention to construct the morphological feature channel, set the multi-head attention parameters, including the number of attention heads and the output dimension of each head, and construct three sets of mapping matrices for each attention head. Perform linear transformation to obtain query Q, key K, value V tensors, perform self-attention calculation on each attention head, and concatenate the output vectors of all heads along the dimension to form the final attention modeling output, which represents the health status representation vector modeled by the current morphological feature channel;

[0124] Receive the output of the time channel and morphological feature channel and splice them to form a fusion vector , construct the MLP network structure, including input layer, hidden layer and output layer, Input into MLP to get a preliminary SOH estimate .

[0125] Through a feature-partitioning modeling strategy, the team fully decoupled temporal dynamics from frequency-domain morphological signals, improving modeling accuracy. EMD and multi-scale feature structures were introduced to enhance the representation of non-stationary signals. APSO was combined with LSTM structural optimization to address the weak stability of traditional deep network architectures. The Attention mechanism was used to model the inherent dependencies between high-order statistical features, enhancing the adaptability of SOH estimation to complex aging characteristics. This approach effectively improved the accuracy and generalization of static SOH estimation, providing strong modeling support for the automated and intelligent screening of retired batteries.

[0126] Furthermore, the joint extended Kalman filter is used to perform SOH correction, and the current equivalent internal resistance estimation is obtained including:

[0127] Collect battery factory rated capacity , calculate the initial capacity estimate based on the SOH estimate and rated capacity :

[0128]

[0129] Construct and initialize the filter state vector, which includes the current state of charge, RC voltage term (electrochemical polarization voltage), current equivalent internal resistance, and current battery capacity (the estimated capacity is used as the initial value of the current battery capacity);

[0130] At the same time, the covariance matrix, state noise and observation noise are initialized based on historical data;

[0131] Extract dynamic observation data sequences from the standard charging test process, including current, terminal voltage, and time intervals. Align and remove anomalies in the data. Use the battery RC equivalent model to predict state variables, including the state of charge and RC voltage terms:

[0132]

[0133]

[0134] Where, is the state of charge at the kth moment, is the estimated capacity at the kth moment, is the current, is the sampling time interval, is the polarization resistance, is a polarized capacitor, is the state of charge at the next moment, is the RC polarization voltage at the next moment, is the RC polarization voltage;

[0135] The capacity and internal resistance remain unchanged during the state prediction (considered as constants during the prediction phase);

[0136] Calculate the current predicted voltage based on the state prediction value :

[0137]

[0138] Where, is the open circuit voltage value, obtained by experimental fitting or OCV-SOC table lookup, is the ohmic internal resistance;

[0139] Calculate the error between the actual voltage and the predicted voltage :

[0140]

[0141] Where, is the actual measured voltage value;

[0142] Use the standard EKF Kalman gain formula to update the state vector and state covariance:

[0143]

[0144]

[0145]

[0146] Where, is the Kalman gain, O is the observation noise covariance matrix, is the Jacobian matrix of the observation function, is the updated state vector, is the state covariance forecast, is the transposed matrix, is the current predicted state, is the updated state covariance matrix, Is the identity matrix, used to keep the matrix dimension unchanged;

[0147] After all sampling points are iterated, the final state vector is estimated, including the current state of charge estimate, RC voltage term estimate, current equivalent internal resistance estimate, and current battery capacity estimate.

[0148] Extract the battery's current capacity estimate from the final state vector estimate and calculate the final health estimate using the battery's current capacity estimate and rated capacity. :

[0149]

[0150] Where, is the final estimated capacity.

[0151] In terms of state modeling, the construction of the joint state vector not only incorporates the battery's real-time state of charge (SOC) and RC voltage term, but also incorporates two core parameters: the current equivalent internal resistance and capacity. This upgrades the SOH estimation process from a simple "voltage-capacity mapping" problem to a multi-parameter co-evolution modeling problem. This joint state-parameter filtering structure improves the model's responsiveness to internal system state perturbations, making SOH estimation more adaptable and real-time. During the prediction phase, the battery's dynamic response is accurately modeled using an RC equivalent circuit model, and the SOC and polarization voltage are calculated. The predicted voltage is then calculated using the OCV-SOC mapping function, effectively reproducing the terminal voltage behavior under actual operating conditions. By introducing observation residuals through error calculation with the measured voltage, the system dynamically adjusts the state estimate. State correction through Kalman gain calculation ensures high estimation stability and accuracy in the face of uncertainties such as voltage noise and load disturbances. This feedback-based estimation mechanism is crucial for improving the robustness of SOH output.

[0152] S4: Estimate the maximum available power of the battery based on the current equivalent internal resistance estimate and real-time voltage , based on SOH and Classify battery health levels and recommend application scenarios;

[0153] Specifically, the maximum available power of the battery is estimated based on the current capacity estimation and real-time voltage , based on SOH and Battery health level classification and application scenario recommendations include:

[0154] Extract the current equivalent internal resistance estimate from the final state vector estimate , and estimate the maximum available power of the battery based on the real-time voltage U :

[0155]

[0156] The final health estimate and the maximum available power of the battery are integrated to generate an input feature vector. A random forest is used as a classification model, and health level labels are defined as A, B, and C, corresponding to high health, medium health, and low health, respectively. The input feature vector is input into the random forest model to obtain the corresponding health level labels.

[0157] Based on the classification results, the corresponding application scenarios are automatically recommended, including Class A recommendations for industrial energy storage systems, grid backup power supplies, microgrid main supply systems, data center UPS, etc., Class B recommendations for low-speed electric vehicles, household small energy storage systems, emergency communication and power supply, etc., and Class C recommendations for material recycling, regeneration and disassembly, and battery cell cascade remanufacturing, etc.

[0158] By combining current internal resistance and real-time voltage to model power capability, this approach overcomes the limitations of traditional single-dimensional estimations based solely on open-circuit voltage or internal resistance. By incorporating the SOP factor into the health grading, a joint feature vector is constructed, integrating the battery's "remaining life" and "current performance" into the classification process, improving grading granularity and application-matching accuracy. When processing multidimensional numerical features (such as SOH, SOP, feature subset mean, and skewness), the random forest algorithm automatically assesses feature importance and eliminates redundant features, making classification results more stable and interpretable. The classification results are matched to actual application scenarios, directly breaking down the information barriers between the assessment and application sides. For example, Grade A batteries are recommended for industrial and commercial energy storage or microgrid systems, Grade B batteries are suitable for light-duty mobility vehicles, and Grade C batteries are included in recycling and disposal pathways. This significantly improves the efficiency of implementing assessment results and helps promote the development of a closed-loop battery recycling industry chain.

[0159] S5. Package all data into an assessment report and archive it in the cloud;

[0160] Specifically, encapsulating all data into an assessment report and archiving it in the cloud means integrating the basic battery information and the data generated by the analysis (health level) according to the task number and the bound data, generating a complete battery health assessment report, outputting the report in both PDF and JSON structured data formats, uploading the report to the cloud database platform, connecting it to the battery life cycle management system (BMS), and marking the corresponding battery life cycle status as "assessed".

[0161] It has achieved the organic unity of visualization of assessment results, improvement of data availability and intelligent linkage of the system, and significantly improved the industrial adaptability, data reusability and management intelligence level of the retired battery health assessment system.

[0162] This embodiment also provides a health assessment system for retired batteries, including:

[0163] The task management module is used to identify the retired battery identification and automatically generate an assessment task number, setting the battery status to "pending inspection";

[0164] Feature extraction module, used to set the battery charging mode, collect battery status data in real time and extract battery features to construct feature vectors;

[0165] The estimation module is used to input the feature vector into the parallel attention neural network for static SOH estimation to obtain a preliminary SOH estimate;

[0166] A correction module is used to perform SOH correction using a joint extended Kalman filter and obtain the current equivalent internal resistance estimate;

[0167] The grading module estimates the maximum available power of the battery based on the current equivalent internal resistance and real-time voltage, and performs health grading and application scenario recommendations based on SOH.

[0168] The archiving module is used to package all data into assessment reports and archive them in the cloud.

[0169] This embodiment also provides a computer device suitable for use in a health assessment method for retired batteries, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the health assessment method for retired batteries proposed in the above embodiment.

[0170] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0171] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the health assessment method for retired batteries proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0172] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A health assessment method for retired batteries, characterized by: include, Identify the retired battery logo and automatically generate an assessment task number, setting the battery status to "pending inspection"; Set the battery charging mode, collect battery status data in real time and extract battery features to construct feature vector G; The battery status data includes voltage, current and time data; The feature vector G is input into the parallel attention neural network for static SOH estimation to obtain a preliminary SOH estimate, and the joint extended Kalman filter is used to correct the SOH and obtain the current equivalent internal resistance estimate; Estimate the maximum available power of the battery based on the current equivalent internal resistance estimate and real-time voltage , based on SOH and Classify battery health levels and recommend application scenarios; Package all data into an assessment report and archive it in the cloud; The inputting of the feature vector G into the parallel attention neural network to perform static SOH estimation and obtain a preliminary SOH estimation value includes: Divide the feature vector G into two sub-vectors according to the feature source, including the time series feature vector and morphological feature vectors , respectively used for different model channel processing; from The main variables reflecting the evolution of battery charge characteristics are extracted and recorded as signal sequence x(t). The empirical mode decomposition algorithm is used to decompose the signal sequence x(t) into several intrinsic mode functions and residuals. Select the first two order IMF components and Used to reconstruct the main sequence ; Combine the features of all time steps into the input tensor , where t is the time step and d is the number of input feature dimensions at each time step; Use the LSTM model to build the time channel, build the first layer of LSTM-1, initialize the input dimension, number of hidden units, activation function, and input Use a linear layer to increase the dimension to the same size as the residual dimension and normalize the output vector at each time step; The normalized output vector is fed into the second layer LSTM-2 network, and the input Use a linear layer to increase the dimension to the same dimension to match the residual dimension, normalize the output vector of each time step, perform a time average pooling operation on the output vector, send the pooled vector to the fully connected network, and output it by the output layer; Collect historical battery data as a training set and input it into the LSTM model for iterative training. Define the mean square error as the training target loss function and use the Adam optimizer to iteratively optimize the model parameters. If the error does not decrease after P consecutive rounds of verification, stop the iteration and obtain the optimal model parameters as the initial parameter set E. Expand E into a one-dimensional vector as the initial position of the particle and initialize the APSO group, set the group size, and initialize a velocity vector for each particle at the same time; Start iterative optimization, set the maximum number of rounds, perform speed and position updates for each particle, calculate the fitness value corresponding to each particle, and update the individual optimal solution and the global optimal solution; After completing the global search iteration of APSO, the current global optimal particle position is directly extracted as the final optimization solution and restored to the LSTM model parameter structure. The optimized parameters are loaded into the original LSTM model to obtain the optimized model; Will Input the optimized LSTM model to obtain the preliminary SOH prediction value; Use Attention to construct the morphological feature channel, set the multi-head attention parameters, including the number of attention heads and the output dimension of each head, and construct three sets of mapping matrices for each attention head. Perform linear transformation to obtain query Q, key K, value V tensors, perform self-attention calculation on each attention head, and concatenate the output vectors of all heads along the dimension to form the final attention modeling output, which represents the health status representation vector modeled by the current morphological feature channel; Receive the output of the time channel and morphological feature channel and splice them to form a fusion vector , construct the MLP network structure, including input layer, hidden layer and output layer, Input into MLP to get a preliminary SOH estimate .

2. The health assessment method for retired batteries according to claim 1, wherein: Setting the battery charging mode, collecting battery status data in real time and extracting battery features to construct a feature vector G refers to collecting battery voltage, current and time data and recording them as an original data set; After preprocessing the data in the original data set, the primary features of the data are extracted, including the charging time in the constant current stage , Charging time in constant voltage stage , voltage mid-segment rise rate , incremental capacity main peak , Corresponding voltage position and the peak time point of the differential voltage curve ; Integrate the extracted features to form a first-level feature set ; The voltage and current in the data set are decomposed into three layers of wavelet using Daubechies4 wavelet basis function to obtain six groups of sub-signals. For each sub-signal, features are extracted and integrated to generate a wavelet feature set. ; The feature set and Merge and construct the total feature set F; Perform a full discharge on each battery to obtain the true capacity label a. Use the Pearson correlation evaluation method to calculate the correlation between each dimension feature and the true capacity label. Retain the features with correlation greater than the preset threshold and integrate them into the filtered feature set F'. Construct a binary selection variable for each feature in F' to construct a binary vector of length k ,in: Indicates whether the kth feature in the full feature set is selected, and each B corresponds to a feature subset combination; Use the sparrow search algorithm to search for the optimal feature subset, and divide the individual roles into the first The role of the individual is the discoverer, and the remaining individuals are followers; Based on the feature data, the XGBoost regressor is used to obtain the predicted capacity, and the average error between the predicted capacity and the actual capacity is calculated as the fitness value. The individual with the lowest fitness is set as the global optimal solution. ; For the discoverer, an exponential decay perturbation update strategy is implemented; Perform global guidance adjustment on the remaining individuals with reference to the current global optimal solution; The individual with the highest current fitness is marked as a vigilant, and mutation perturbation is performed to escape the local optimum; The continuous coding vectors of all individuals are probability mapped through the Sigmoid function, and then thresholded and binarized to obtain the updated binary coding individuals. After reaching the maximum number of iterations W, the final global optimal coding solution A is output. According to the optimal coding A, the selected feature dimensions are extracted to form the optimal feature subset, which is normalized and then spliced into the feature vector G.

3. The health assessment method for retired batteries according to claim 2, wherein: The method of using the joint extended Kalman filter to correct the SOH and obtain the current equivalent internal resistance estimate includes: Collect battery factory rated capacity , calculate the initial capacity estimate based on the SOH estimate and rated capacity; Construct and initialize the filter state vector, which includes the current state of charge, RC voltage term, current equivalent internal resistance, and current battery capacity; At the same time, the covariance matrix, state noise and observation noise are initialized based on historical data; Extract dynamic observation data sequences from the standard charging test process and use the battery RC equivalent model to predict the filtered state vector, including the current state of charge and RC voltage term. Calculate the current predicted voltage based on the predicted value of the filtered state vector ; Calculate the error between the actual voltage and the predicted voltage; Update the state vector and state covariance using the standard EKF Kalman gain formula; After all sampling points are iterated, the final state vector is estimated, including the current state of charge estimate, RC voltage term estimate, current equivalent internal resistance estimate, and current battery capacity estimate. Extract the battery's current capacity estimate from the final state vector estimate and calculate the final health estimate using the battery's current capacity estimate and rated capacity. .

4. The health assessment method for retired batteries according to claim 3, wherein: The maximum available power of the battery is estimated based on the current equivalent internal resistance estimation and real-time voltage , based on SOH and Battery health level classification and application scenario recommendations include: Extract the current equivalent internal resistance estimate from the final state vector estimate , and estimate the maximum available power of the battery based on the real-time voltage U ; The final health valuation and the maximum available power of the battery are integrated to generate an input feature vector. A random forest is used as the classification model, and health level labels are defined as A, B, and C, corresponding to high health, medium health, and low health, respectively. The input feature vector is input into the random forest model to obtain the corresponding health level label, and the corresponding application scenario is automatically recommended based on the classification results.

5. The health assessment method for retired batteries according to claim 4, characterized in that: The identification of retired battery identification and automatic generation of assessment task number, and setting the battery status to "pending detection" means that during the battery warehousing process, the RFID tag information attached to the battery is read by an RFID scanning device and used as the unique number of the battery. After obtaining the battery number, the assessment task number corresponding to the battery is synchronously generated, and the assessment task number and the battery number are bound one by one. After the information is bound, the initial status of the task is set to "pending detection".

6. The health assessment method for retired batteries according to claim 5, characterized in that: The packaging of all data into an evaluation report and archiving and storing it in the cloud refers to integrating the basic information of the battery and the data generated by the analysis, generating a complete battery health evaluation report, and uploading it to the cloud database platform.

7. A health assessment system for retired batteries, based on the health assessment method for retired batteries according to any one of claims 1 to 6, characterized in that: include, The task management module is used to identify the retired battery logo and automatically generate an assessment task number, setting the battery status to "pending inspection"; Feature extraction module, used to set the battery charging mode, collect battery status data in real time and extract battery features to construct feature vectors; The estimation module is used to input the feature vector into the parallel attention neural network for static SOH estimation to obtain a preliminary SOH estimate; A correction module is used to perform SOH correction using a joint extended Kalman filter and obtain the current equivalent internal resistance estimate; The grading module estimates the maximum available power of the battery based on the current equivalent internal resistance and real-time voltage, and performs health grading and application scenario recommendations based on SOH. The archiving module is used to package all data into assessment reports and archive them in the cloud.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the health assessment method for retired batteries according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the health assessment method for retired batteries according to any one of claims 1 to 6 are implemented.

Images