Reliability optimization design method for retired power battery system

Abstract

The invention relates to the technical field of power battery echelon utilization, and discloses a decommissioned power battery system reliability optimization design method, which comprises the following steps: acquiring and preprocessing battery operation attenuation data; constructing an attenuation prediction model; performing dynamic clustering and grouping after multi-dimensional consistency evaluation; and generating a strategy and feeding back iterative optimization. The system comprises a data acquisition module, a preprocessing module, an attenuation prediction module, a consistency evaluation module, a dynamic grouping module and a strategy execution and feedback module. By adopting the scheme, the problems are solved, the grouping accuracy and stability are improved, the secondary utilization efficiency is improved, the cycle life is prolonged, the operation reliability is improved, and the maintenance cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of power battery cascade utilization technology, specifically relating to a reliability optimization design method for retired power battery systems. Background Technology

[0002] With the rapid development of the electric vehicle industry, a large number of power batteries are gradually reaching the end of their design life, forming a massive resource of retired power batteries. These retired batteries have significant value for secondary use in energy storage, backup power, and other fields, and are of great importance for achieving resource recycling, reducing environmental impact, and improving economic efficiency. Efficient and reliable reconfiguration and management are key aspects of the secondary utilization of retired power batteries.

[0003] Among these, the performance evaluation and regrouping of retired power batteries are crucial to ensuring the reliability and efficiency of the cascade utilization system. Existing regrouping methods typically aim to screen and match individual cells to form battery modules or packs with relatively consistent performance. The basic principle is to classify batteries based on key parameters such as voltage, internal resistance, and capacity, in order to maximize the overall performance and lifespan of the regrouped battery pack.

[0004] However, existing technologies have several shortcomings in the reorganization of retired power batteries. First, traditional static grouping methods primarily match batteries based on their performance parameters at a specific moment, failing to effectively address the inherent high dispersion and dynamic degradation characteristics of retired batteries. This leads to a significant "weakest link" effect within the battery pack, where a few rapidly degrading batteries drag down the overall charging / discharging efficiency and lifespan of the entire pack. Second, existing technologies generally lack the ability to effectively predict future battery degradation trends, relying solely on current parameters for grouping. This means that the grouping scheme may fail in the short term due to battery performance differentiation, and frequent adjustments not only increase maintenance costs but also lead to battery disassembly and assembly losses. Furthermore, the performance of retired batteries is influenced by a combination of multiple parameters, while traditional grouping decisions often over-rely on a few core parameters such as voltage and capacity, easily neglecting equally critical performance indicators such as internal resistance and temperature consistency. This results in the reliability and accuracy of the grouping results failing to meet high standards. Therefore, a method that can overcome these shortcomings and achieve reliable optimization design for retired power battery systems is urgently needed. Summary of the Invention

[0005] This invention provides a reliability optimization design method for retired power battery systems, aiming to solve problems in the prior art such as inherent performance inconsistencies, the "weakest link" effect caused by static grouping strategies, lack of prediction of battery degradation trends, and insufficient multi-parameter collaborative decision-making capabilities in retired power batteries. This method aims to improve the overall operating efficiency of retired power battery systems, extend their cycle life, and reduce operating and maintenance costs.

[0006] This invention provides a reliability optimization design method for retired power battery systems, which includes the following steps: Acquire real-time operating data and historical degradation data of the power batteries to be restructured and retired. The real-time operating data includes, but is not limited to, the voltage, current, temperature, internal resistance and state of charge of individual batteries. The historical degradation data includes, but is not limited to, the capacity, health status and cumulative number of cycles of individual batteries. The real-time running data and historical decay data are preprocessed, including data cleaning, missing value imputation, time series alignment and feature normalization. Based on the preprocessed historical degradation data, a single cell degradation trend prediction model is constructed and trained. The model is used to predict the degradation trend of key performance parameters of each single cell in a future preset time period. The key performance parameters include capacity degradation rate and internal resistance growth rate. Based on the real-time operating data, historical degradation data, and predicted degradation trends, a multi-dimensional consistency evaluation index system is constructed, and the comprehensive consistency evaluation score between each individual cell and other individual cells to be recombined is calculated. The multi-dimensional consistency evaluation index system includes, but is not limited to, voltage consistency, capacity consistency, internal resistance consistency, temperature consistency, and future degradation trend consistency. Based on the comprehensive consistency evaluation score, the power batteries to be reorganized and retired are grouped using a dynamic clustering algorithm, so that the comprehensive consistency evaluation score of individual batteries in the same group reaches a preset threshold and the performance difference between groups is maximized. The dynamic grouping strategy is generated and executed, the recombined battery system operating performance data is monitored and recorded, and the operating performance data is fed back to the data acquisition step for iterative optimization.

[0007] In a preferred embodiment of the present invention, the acquisition of real-time operating data and historical degradation data of the retired power battery to be restructured specifically includes: Real-time voltage data is obtained by connecting to the battery management system of each individual cell, with an accuracy of 0.1 millivolts; Real-time current data is acquired using a Hall effect sensor with an accuracy of 0.1 amperes. Real-time temperature data is acquired using an embedded thermistor array with an accuracy of 0.1 degrees Celsius. Real-time internal resistance data were obtained using an AC impedance spectroscopy analyzer with an accuracy of 0.1 milliohms. Real-time state of charge data can be calculated and obtained using the Coulomb integral method or the open-circuit voltage method. Historical capacity decay data is obtained through periodic capacity testing, which employs constant current and constant voltage charging and constant current discharging modes. Accumulated cycle count data can be obtained through a battery management system or an external data logger.

[0008] In a preferred embodiment of the present invention, the data preprocessing specifically includes: The 3σ principle or the local outlier algorithm is used to detect outliers in the collected data, and median filtering or a statistical model based on historical data is used to correct outliers. Missing data points can be filled using linear interpolation or cubic spline interpolation methods; Dynamic time warping algorithms or synchronous sampling techniques are used to align the time series of data from different sensors, unifying all data to a preset sampling frequency. All feature data are normalized using the minimum-maximum normalization method or the Z-score normalization method, mapping the data to a preset numerical range.

[0009] In a preferred embodiment of the present invention, the construction and training of the single-cell battery degradation trend prediction model specifically includes: The preprocessed historical decay data is divided into a training dataset and a validation dataset. A deep learning model based on a temporal convolutional network or a long short-term memory network is constructed. The input layer of the model receives multi-dimensional temporal features of a single battery cell, including historical capacity, historical internal resistance, historical voltage curve features, historical temperature curve features, and cumulative cycle count. The hidden layer contains multiple convolutional layers or recurrent layers to extract temporal dependent features. The output layer outputs the predicted values ​​of capacity decay rate and internal resistance growth rate within a future preset time period. The mean squared error is used as the loss function, and the deep learning model is trained using an adaptive moment estimation optimizer. The model performance is evaluated on the validation dataset until a preset prediction accuracy requirement is met, which is that the mean absolute error is less than a preset threshold.

[0010] In a preferred embodiment of the present invention, the future preset time period is between 24 hours and 72 hours.

[0011] In a preferred embodiment of the present invention, the construction of a multi-dimensional consistency evaluation index system specifically includes: The voltage consistency index is defined as the relative deviation between the current voltage of a single cell and the average voltage of all single cells in the set to be grouped, or the standard deviation of voltage within the set. The capacity consistency index is defined as the relative deviation between the current capacity of a single cell and the average capacity of all single cells in the set to be grouped, or the standard deviation of the capacity of the set. The internal resistance consistency index is defined as the relative deviation between the current internal resistance of a single cell and the average internal resistance of all single cells in the set to be grouped, or the standard deviation of internal resistance within the set. The temperature consistency index is defined as the relative deviation between the current temperature of a single cell and the average temperature of all single cells in the set to be grouped, or the temperature standard deviation within the set. The consistency index of future degradation trend is defined as the relative deviation between the predicted capacity degradation rate of a single cell and the average predicted capacity degradation rate within the set, or the standard deviation of the predicted capacity degradation rate within the set, and the relative deviation between the predicted internal resistance growth rate of a single cell and the average predicted internal resistance growth rate within the set, or the standard deviation of the predicted internal resistance growth rate within the set. The weight coefficients of each consistency index are determined by using the analytic hierarchy process (AHP) or principal component analysis (PCA), and the weighted sum is calculated as the overall consistency evaluation score.

[0012] In a preferred embodiment of the present invention, the dynamic clustering algorithm is the K-means++ algorithm, the DBSCAN algorithm, or the spectral clustering algorithm. The clustering algorithm uses the comprehensive consistency evaluation score as the clustering feature for iterative calculation, so that the individual cells within the same cluster have high similarity, and the individual cells between different clusters have low similarity.

[0013] In a preferred embodiment of the present invention, generating and executing the dynamic grouping strategy specifically includes: Based on the clustering results of the dynamic clustering algorithm, a battery physical reorganization instruction is generated, which specifies the battery pack number to which each individual battery should belong. The instructions are transmitted to an automated battery reconfiguration device, which is responsible for disassembling, transporting, and reinstalling individual batteries to form a new battery pack. Functional tests were conducted on the recombined battery system to verify that its performance met the design requirements.

[0014] This invention provides a reliability optimization design system for retired power battery systems, comprising: The data acquisition module is used to acquire real-time operating data and historical degradation data of the power batteries to be restructured and retired. The real-time operating data includes, but is not limited to, the voltage, current, temperature, internal resistance and state of charge of individual batteries. The historical degradation data includes, but is not limited to, the capacity of each charge and discharge cycle, health status and cumulative number of cycles of individual batteries. The data preprocessing module is used to perform data cleaning, missing value imputation, time series alignment, and feature normalization on the real-time running data and historical decay data. The degradation trend prediction module is used to construct and train a single cell degradation trend prediction model based on the preprocessed historical degradation data, and use the model to predict the degradation trend of key performance parameters of each single cell within a future preset time period. The key performance parameters include capacity degradation rate and internal resistance growth rate. The consistency evaluation module is used to construct a multi-dimensional consistency evaluation index system based on the real-time operating data, historical degradation data and predicted degradation trend, and to calculate the comprehensive consistency evaluation score between each individual cell and other individual cells to be recombined. The multi-dimensional consistency evaluation index system includes, but is not limited to, voltage consistency, capacity consistency, internal resistance consistency, temperature consistency and future degradation trend consistency. The dynamic grouping module is used to group the retired power batteries to be reorganized into groups based on the comprehensive consistency evaluation score using a dynamic clustering algorithm, so that the comprehensive consistency evaluation score of the individual batteries in the same group reaches a preset threshold and the performance difference between groups is maximized. The strategy execution and feedback module is used to generate and execute the dynamic grouping strategy, monitor and record the operating performance data of the recombined battery system, and feed the operating performance data back to the data acquisition module for iterative optimization.

[0015] In a preferred embodiment of the present invention, the data acquisition module includes: Voltage sensors are used to monitor the voltage of individual battery cells in real time. Current sensor used to monitor the current of a single battery cell in real time; Temperature sensor used to monitor the temperature of individual battery cells in real time; Internal resistance testing unit, used to monitor the internal resistance of individual cells in real time; The state of charge calculation unit is used to calculate the state of charge of a single battery cell. The capacity testing equipment interface is used to connect to external capacity testing equipment to obtain historical capacity degradation data; The data storage unit is used to store all collected real-time operating data and historical attenuation data.

[0016] In a preferred embodiment of the present invention, the data preprocessing module includes: An outlier detection unit is used to identify and correct outliers in the data; Missing value imputation unit, used to fill in missing values ​​in the data; Time alignment units are used to align time series data from different sources; The normalization processing unit is used to normalize data features.

[0017] In a preferred embodiment of the present invention, the attenuation trend prediction module includes: Model building unit, used to build deep learning models based on temporal convolutional networks or long short-term memory networks; A model training unit is used to train the deep learning model using historical decay data. The prediction output unit is used to output the predicted values ​​of capacity decay rate and internal resistance growth rate of each individual cell over a future preset time period.

[0018] In a preferred embodiment of the present invention, the consistency evaluation module includes: The indicator definition unit is used to define indicators such as voltage consistency, capacity consistency, internal resistance consistency, temperature consistency, and future degradation trend consistency. The weight calculation unit is used to determine the weight coefficients of each consistency index; The comprehensive score calculation unit is used to calculate the comprehensive consistency evaluation score for each individual cell.

[0019] In a preferred embodiment of the present invention, the dynamic grouping module includes: The clustering algorithm execution unit is used to perform K-means++ algorithm, DBSCAN algorithm or spectral clustering algorithm to cluster individual cells; The grouping result generation unit is used to generate grouping results to ensure that the overall consistency evaluation score of individual cells within the same group reaches a preset threshold.

[0020] In a preferred embodiment of the present invention, the strategy execution and feedback module includes: The instruction generation unit is used to generate battery physical recombination instructions based on the grouping results. An automated battery rebuilding interface is used to communicate with automated battery rebuilding equipment and transmit the rebuilding instructions; The performance monitoring unit is used to monitor the operating performance data of the reconstituted battery system; The data feedback unit is used to feed back the operating performance data to the data acquisition module.

[0021] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. This invention solves the problem that traditional static grouping methods cannot predict the future degradation state of batteries by introducing a single-cell degradation trend prediction model. By identifying single cells with high degradation risk in advance, frequent regrouping caused by short-term performance differentiation can be avoided, reducing the grouping adjustment frequency from twice a day to once a day, thereby reducing battery disassembly and assembly losses and related maintenance costs.

[0022] 2. This invention overcomes the limitations of existing technologies that rely solely on a single or few parameters (such as voltage and capacity) for grouping by constructing a multi-dimensional consistency evaluation index system and combining it with a dynamic clustering algorithm. This method comprehensively considers key operating parameters such as voltage, capacity, internal resistance, and temperature, as well as the most important future degradation trend, thereby effectively improving the accuracy and stability of grouping and avoiding the "weakest link" effect.

[0023] 3. The dynamic clustering grouping strategy adopted in this invention can adaptively adjust according to the real-time performance data of the battery and the predicted degradation trend. This dynamic optimization strategy can maximize the secondary utilization efficiency of retired power batteries, extend their overall cycle life, and improve the operational reliability of the entire battery system.

[0024] 4. This invention adopts a closed-loop optimization mechanism, which continuously monitors the operating performance data of the recombined battery system and feeds it back to the data acquisition stage for iterative optimization. This ensures that the system can maintain the optimal grouping strategy for a long time and adapt to the dynamic changes in battery performance, thereby achieving efficient, stable and long-life operation of the retired power battery system. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the core principle framework of the dynamic clustering and grouping strategy based on the prediction of the degradation trend of individual cells in this invention. Figure 2 This is a logical flowchart of data preprocessing and feature construction in this invention; Figure 3 This is a flowchart illustrating the logical framework for constructing and training the single-cell battery degradation trend prediction model in this invention. Figure 4 This is a flowchart illustrating the execution and closed-loop optimization process of the dynamic grouping strategy in this invention. Detailed Implementation

[0026] Please refer to Figure 1-4This invention proposes a reliability optimization design method for retired power battery systems. By introducing a dynamic clustering grouping strategy based on the predicted degradation trends of individual battery cells, intelligent management and optimized configuration of retired power batteries are achieved. The core of this method lies in constructing an accurate degradation trend prediction model through real-time collection and in-depth analysis of multi-dimensional battery operating data. This model identifies batteries with high-risk degradation trends before individual battery performance differences emerge. Based on this, a dynamic clustering algorithm is used to group the batteries, combining real-time performance parameters and predicted degradation states of individual cells, ensuring high consistency in current performance and future degradation trends across each group. Furthermore, this method introduces a closed-loop optimization mechanism. By continuously monitoring the battery pack's operating performance, the prediction model and grouping strategy are iteratively updated to adapt to long-term dynamic changes in the batteries, thereby improving the overall reliability, efficiency, and cycle life of the retired power battery system.

[0027] As one embodiment of the present invention, the reliability optimization design method for retired power battery systems includes the following steps: S101 collects and preprocesses the operating data and historical degradation data of individual cells in the retired power battery system.

[0028] In this embodiment, step S101 aims to acquire raw data for evaluating the current state of individual cells, predicting their future degradation trends, and supporting dynamic grouping decisions, and transform it into a structured, high-quality set of usable features. This process is a fundamental step in the entire optimization design method, ensuring that subsequent prediction and clustering algorithms receive accurate, comprehensive, and representative input information.

[0029] The data acquisition process is achieved through a series of high-precision sensors and data acquisition modules integrated into the retired power battery system and its testing platform. These sensors include, but are not limited to, voltage sensors, current sensors, temperature sensors, and internal resistance testers. Voltage sensors monitor the terminal voltage of each individual battery cell in real time, with a sampling frequency set to 100 Hz to capture transient voltage changes during charging and discharging. Current sensors monitor the charging and discharging current flowing through each individual battery cell, with a sampling frequency synchronized with the voltage sensors to ensure accurate alignment of current and voltage data. Temperature sensors are typically located on the surface or inside the individual battery cells to monitor the battery's operating temperature, with a sampling frequency of 10 Hz. Internal resistance testers periodically measure the AC or DC internal resistance of individual batteries under specific test conditions, such as every 24 hours, to reflect changes in the ohmic and electrochemical polarization states within the battery. Furthermore, charge and discharge capacity data of the battery needs to be collected periodically through coulomb counting or constant current charge and discharge testing, for example, a capacity test after every 100 charge and discharge cycles, to quantify the actual usable energy of the battery. All collected raw data is precisely timestamped and transmitted to the data storage and management server via the battery management system's data bus, such as the controller area network bus or Ethernet interface. The data storage and management server adopts a distributed time-series database architecture to ensure efficient storage of massive amounts of multi-dimensional, high-frequency time-series data and supports fast querying and indexing.

[0030] Furthermore, the preprocessing process includes multiple sub-steps designed to improve data quality and usability.

[0031] First, data cleaning is performed to identify and correct outliers and noise in the raw data. Outlier detection employs statistical methods, such as the three-standard-deviation criterion or the local anomaly factor algorithm. For identified outlier data points, different processing strategies are adopted based on their causes: if caused by a momentary sensor malfunction, mean interpolation or cubic spline interpolation of adjacent time points is used for replacement; if it is a persistent anomaly, the battery or sensor is marked as faulty, and a maintenance alarm is triggered. Noise filtering uses digital filters, such as Kalman filters or moving average filters, to smooth the data and remove high-frequency random noise.

[0032] Secondly, data synchronization and alignment are performed. Since different sensors may have different sampling frequencies, and data transmission may introduce time delays, all data needs to be synchronized and aligned according to a unified time base. This process is achieved through timestamp correction and resampling techniques, unifying all data to a granularity of one hertz per second, ensuring consistency across different data types in the time dimension.

[0033] Next, missing value imputation is performed. When sensor data is missing for various reasons, data imputation techniques are used to fill the gaps. For short-term missing data, linear or polynomial interpolation is used; for long-term missing data, predictive imputation is performed by combining historical data trends or adjacent battery data.

[0034] Subsequently, data normalization or standardization is performed. To eliminate the impact of data with different dimensions on the performance of subsequent algorithms, all numerical features are normalized, scaling them to the range of zero to one, or standardized so that their mean is zero and their variance is one. For example, the voltage value of a single battery cell can be normalized to the range of zero to one, as can the internal resistance value of a single battery cell.

[0035] Finally, feature construction is performed. High-level, physically meaningful features are extracted from the preprocessed raw data. These features include, but are not limited to, the slope of the charge / discharge voltage curve of a single cell, the instantaneous rate of change of internal resistance, the temperature gradient, the capacity decay rate, and the energy efficiency under a specific charge / discharge cycle. In addition, time-series statistical features can be constructed, such as the variance of voltage fluctuations, the average temperature, and the periodic variation amplitude of internal resistance. These features are generated through aggregation, transformation, and derivation operations, aiming to more comprehensively and accurately characterize the health status and degradation behavior of a single cell. For example, the capacity decay rate is calculated by comparing the percentage change between the current capacity and the initial capacity; the impedance spectrum characteristics of the battery are obtained through frequency domain analysis of the internal resistance data. All constructed features are organized in the form of multi-dimensional feature vectors, with each vector representing the overall performance state of a single cell at a specific time point.

[0036] S102, Based on the preprocessed data, construct and train a single-cell battery degradation trend prediction model.

[0037] In this embodiment, step S102 aims to establish a mathematical model capable of accurately predicting the future performance degradation state of individual battery cells. Traditional methods often group cells based solely on current performance data, lacking the ability to predict the dynamic degradation process within the battery's lifespan, thus resulting in insufficient short-term effectiveness of the grouping scheme. This step overcomes this limitation by introducing an advanced time-series prediction model, providing crucial future state information for subsequent dynamic clustering grouping strategies.

[0038] The model building and training process includes the following core steps.

[0039] First, a suitable model architecture is selected. Given the temporal dependence and nonlinear characteristics of the degradation process of a single battery cell, the degradation trend prediction model employs a deep learning architecture, such as a Long Short-Term Memory network or a gated recurrent unit network, or even further, a transformer network based on an attention mechanism. These network structures excel at handling long-term series data and can capture the complex evolution of battery performance parameters over time. The model architecture is defined as a sequence-to-sequence structure with multiple encoder and decoder layers. Each encoder layer includes a multi-head self-attention mechanism and a feedforward network to capture dependencies and nonlinear features at different time scales.

[0040] Secondly, the training dataset is prepared. The training dataset consists of the multi-dimensional feature vector sequence of individual batteries preprocessed and constructed in step S101. Each training sample includes a sequence of battery features within a historical time window as input, such as the daily average voltage, capacity, internal resistance, temperature, and rate of change over the past thirty days, and a target degradation state within a future time window as output, such as the percentage of capacity degradation, internal resistance increment, and voltage plateau decrease after the next seven or thirty days. The training dataset also includes historical degradation trajectories of different retired batteries under different operating conditions to ensure the model's generalization ability. The training dataset is partitioned using a time-series cross-validation method to avoid data leakage and accurately reflect the model's generalization performance. For example, the total dataset is divided into a training set, a validation set, and a test set in chronological order. The training set is used for learning model parameters, the validation set is used for hyperparameter tuning and model selection, and the test set is used for final performance evaluation.

[0041] Subsequently, model training is performed. Model training is conducted through an iterative optimization process. The prepared training dataset is input into the model, and the model's internal parameters are adjusted using the backpropagation algorithm to minimize the error between the predicted values ​​and the actual decay states. The error metric uses mean squared error or mean absolute error as the loss function. The optimizer employs an adaptive learning rate optimization algorithm, such as adaptive moment estimation or adaptive gradient algorithm, to accelerate convergence and improve training stability. During training, appropriate batch sizes and training epochs are set. After each training epoch, the model's performance is evaluated on the validation set, and the learning rate is adjusted based on changes in the validation loss, or an early stopping strategy is employed to prevent overfitting.

[0042] Furthermore, the model training also includes feature importance analysis and model interpretability enhancement. Through interpretable AI techniques, such as attention weight analysis or locally interpretable but unexplained models, the contribution of different input features to degradation prediction is analyzed, thereby deepening the understanding of battery degradation mechanisms and guiding subsequent feature engineering optimization.

[0043] Finally, model evaluation and validation are performed. The trained model is evaluated on an independent test set. Evaluation metrics include prediction accuracy (e.g., the mean absolute error between the predicted capacity decay rate and the actual decay rate), prediction stability, and recall and accuracy for identifying high-risk decaying batteries. Only when the model meets the preset performance requirements is it deployed to subsequent prediction stages.

[0044] S103, using the degradation trend prediction model, predict the future degradation state and risk level of a single cell.

[0045] In this embodiment, step S103 is the practical application of the degradation trend prediction model trained in step S102, aiming to provide a forward-looking decision-making basis for dynamic clustering. By predicting the performance degradation of individual cells over a period of time, the short-sightedness of traditional static grouping can be effectively overcome, avoiding rapid performance differentiation of battery packs due to unforeseen degradation.

[0046] This step is performed periodically, for example, by making a prediction for all individual cells to be grouped every 24 hours or every 12 hours.

[0047] First, real-time input features are acquired. For each individual cell, the latest multi-dimensional feature vector sequence is extracted from the real-time data stream acquired and preprocessed in step S101. This feature vector sequence maintains the same input feature format as used during model training and covers key performance indicators such as voltage, current, temperature, internal resistance, capacity, and their rate of change within the current and past specific time windows. This real-time feature data is transmitted to the prediction model module via a data interface.

[0048] Next, model inference is performed. The real-time input feature sequence is input into the degradation trend prediction model constructed and trained in step S102. The prediction model performs forward propagation calculation on the input sequence and outputs the predicted degradation state of each individual battery cell in a preset future time window, such as the next seven days or the next thirty days. The predicted degradation state includes, but is not limited to, the future capacity degradation percentage, internal resistance increment, voltage plateau drift, and battery health state percentage.

[0049] Subsequently, the risk level is calculated. Based on the predicted degradation state, the risk level of each individual cell is assessed. The risk level is determined based on predefined thresholds. For example, if the predicted future capacity degradation percentage exceeds 10%, or the internal resistance increase exceeds 15%, the individual cell is marked as a high-risk degradation cell; if the degradation is within a preset range, it is marked as a medium-risk degradation cell; and if the degradation is below the threshold, it is marked as a low-risk degradation cell. The risk level assessment can also comprehensively consider multiple degradation indicators, using methods such as weighted summation or decision trees for a comprehensive determination.

[0050] Finally, the prediction results are output. These results are stored in a structured data format, including a unique identifier for each individual battery cell, the predicted degradation state (specific numerical value), and the corresponding risk level. These results serve as core inputs to the subsequent dynamic clustering grouping strategy module, guiding the refined grouping of batteries. This prediction process improves the ability to identify high-risk degradation batteries, identifying potential problem batteries that traditional methods fail to detect in time, thus providing prior information for optimizing the grouping scheme and effectively reducing the frequency of grouping adjustments caused by sudden battery performance degradation.

[0051] S104, Based on the future decay state and risk level, as well as multi-dimensional performance parameters, execute a dynamic clustering grouping strategy.

[0052] In this embodiment, step S104 is a key step in solving the problem of the traditional static grouping bottleneck effect and multi-parameter collaborative decision-making. By fusing real-time performance data of individual cells with the future degradation trend predicted in step S103, and using a dynamic clustering algorithm, the battery pack is finely and proactively grouped, ensuring a high degree of consistency among cells within the group.

[0053] The execution of the dynamic clustering grouping strategy includes the following core steps.

[0054] First, a feature vector for each individual battery is constructed. For each individual battery to be grouped, a high-dimensional feature vector is constructed by combining the real-time multi-dimensional performance parameters obtained in step S101 (e.g., current capacity, internal resistance, open-circuit voltage, cycle count, and temperature consistency index) with the predicted future degradation state and risk level in step S103. The dimensions of the feature vector cover the battery's current electrochemical characteristics, thermal characteristics, and future performance evolution trends. For example, the feature vector may include: current capacity percentage, current internal resistance, average discharge plateau voltage, peak temperature, predicted capacity degradation rate over the next seven days, predicted internal resistance increment over the next thirty days, and numerical encoding of the risk level. To ensure a fair contribution of each dimension of features to the clustering results, the feature vector undergoes a uniform scaling transformation during construction, such as standardization.

[0055] Secondly, a similarity metric and weighting mechanism are defined. Traditional clustering methods often use a single distance metric, which is insufficient to fully reflect the relative importance of multi-dimensional features. This method uses weighted Euclidean distance or weighted cosine similarity as the metric for similarity between individual cells. The weighting mechanism is implemented by assigning a weight value to each feature dimension, reflecting the importance of that feature to the consistency of cell grouping. For example, the weights of future degradation state and core parameters such as current capacity and internal resistance can be set to higher values, while the weights of other auxiliary parameters can be appropriately reduced. These weight values ​​can be set through expert experience or optimized through machine learning algorithms, such as principal component analysis or feature selection algorithms, based on historical grouping results. The formula for calculating the weighted Euclidean distance is expressed as: ; in, and These represent the M-dimensional eigenvectors of two individual cells. and These represent the eigenvectors at the 1st, 2nd, and 3rd eigenvalues, respectively. One portion, This represents the weight of the k-th feature.

[0056] Subsequently, a dynamic clustering algorithm is executed. This method employs a dynamic clustering algorithm based on iterative optimization, such as a variant of the K-means clustering algorithm or the spectral clustering algorithm. In each clustering iteration, this algorithm considers not only the static similarity between batteries but also dynamically adjusts based on the degradation risk predicted in step S103. During algorithm initialization, K cluster centers are randomly selected. In each iteration, each individual battery is assigned to the cluster containing its nearest cluster center; then, the cluster centers of each cluster are recalculated, and the clustering results are updated. The dynamism is reflected in the fact that the clustering algorithm adjusts the calculation of cluster centers and battery assignment based on the risk level of individual batteries. For example, if two batteries are very similar in their current performance parameters, but one is predicted to be a high-risk degradation battery, its distance from healthy batteries will be appropriately increased during the clustering process to avoid assigning it to the same cluster, thereby isolating potential inconsistencies in advance. The clustering process continues iterating until the cluster centers no longer change or the preset maximum number of iterations is reached.

[0057] Furthermore, the effectiveness of the clustering results is evaluated and optimized. To ensure the reliability of the clustering results, metrics such as the silhouette coefficient, Davidson-Bourdin index, or internal and external consistency measures are used to evaluate the clustering effect. If the evaluation results do not meet the preset standards, such as the silhouette coefficient being below a certain threshold, the clustering parameters are adjusted, such as the number of clusters K or feature weights, and the clustering process is repeated until satisfactory clustering results are obtained. The number of clusters K can be determined using the elbow rule or based on domain knowledge.

[0058] Finally, the dynamic clustering grouping results are output. These results are presented as battery pack configuration schemes, clearly indicating the optimal group to which each individual cell belongs. This ensures that the individual cells within each battery pack are highly consistent in terms of current performance parameters and future degradation trends, thereby minimizing the "weakest link" effect. This strategy significantly improves the reliability and accuracy of grouping, far exceeding traditional methods.

[0059] S105, Based on the dynamic clustering grouping results, optimize and adjust the battery pack configuration of the retired power battery system.

[0060] In this embodiment, step S105 is the practical implementation of the dynamic clustering results from step S104, aiming to transform the optimized grouping scheme into specific battery system configuration instructions. This step serves as a bridge between software decision-making and hardware implementation, ensuring that retired power battery systems can be physically or logically reorganized according to the optimal scheme to improve overall reliability and performance.

[0061] The optimization and adjustment process includes the following core steps.

[0062] First, the dynamic clustering grouping results are received and parsed. A data structure containing a unique identifier for each individual battery cell and its optimal group number is received from the dynamic clustering module in step S104. This data structure clearly defines the affiliation of each individual battery cell in the new battery pack configuration. For example, if twenty individual batteries are divided into four groups, then each battery cell will have a new group number next to its identifier.

[0063] Secondly, battery pack reconfiguration instructions are generated. Based on the parsed grouping results, the system automatically generates detailed battery pack reconfiguration instructions. These instructions include: a list of individual cells requiring physical disassembly and reinstallation, the new target location for each individual cell or a new battery pack number, and corresponding connection diagrams or operating procedures. For example, if some cells in an existing battery system need to be moved from one group to another, or completely replaced, the instructions will explicitly specify these operations. For scenarios where physical reconfiguration is not required, such as virtual grouping or dynamic scheduling through an intelligent battery management system, corresponding control parameter update instructions are generated to logically adjust the battery's charging and discharging strategies and energy management.

[0064] Subsequently, physical or logical adjustments to the battery pack configuration are performed. If physical reconfiguration is required, an automated robotic arm or a human operator precisely removes and reconnects the batteries according to reconfiguration instructions. This process ensures that each individual battery cell is correctly placed in its corresponding battery pack according to the new grouping scheme and correctly connected to the data bus and power circuits of the battery management system. To reduce removal and installation losses, the system recommends reconfiguration schemes that minimize the number of physical movements. If logical adjustments are required, configuration update instructions are issued through the battery management system or system-level controller. These instructions update parameters in the battery management system regarding individual battery cell network structure, charge / discharge balancing strategies, fault diagnosis thresholds, etc., achieving virtual grouping optimization without physical intervention. For example, through software configuration, individual batteries with the same degradation trend and performance indicators can be logically divided into a better sub-module for collaborative management.

[0065] Furthermore, the adjusted system functions are verified. After completing the battery pack configuration adjustment, a series of system-level functional and performance tests are performed to verify whether the new battery pack configuration works as expected. These tests include, but are not limited to: overall battery pack charge-discharge cycle testing, equalization performance testing, internal resistance consistency testing, and temperature distribution testing. The effectiveness of the optimization is confirmed by comparing the performance data with that before the adjustment. For example, the overall capacity decay rate, internal resistance drift, and operating temperature consistency of the new battery pack should be better than before the adjustment.

[0066] Finally, update the battery system configuration information. The new battery pack configuration information, including the physical location of each individual cell, its corresponding battery pack number, and the relevant connection topology, is written into the battery management system's configuration database. This information will serve as an important reference for subsequent battery monitoring, fault diagnosis, and closed-loop optimization strategies.

[0067] S106, monitor the actual operating data of the optimized battery pack, and perform closed-loop iterative optimization of the single cell degradation trend prediction model and dynamic clustering grouping strategy.

[0068] In this embodiment, step S106 is the key mechanism for achieving long-term reliability management and continuous performance improvement of the retired power battery system. By constructing a closed-loop feedback system, the actual operating effect of the battery pack is compared with the prediction and grouping strategy, and this feedback information is used to continuously improve the accuracy of the model and the effectiveness of the strategy, thereby ensuring that the entire optimization design method can adapt to the dynamic changes of the battery during its service life.

[0069] The closed-loop iterative optimization process includes the following core steps.

[0070] First, the actual operating data of the battery pack is continuously monitored. After the battery pack configuration optimization and adjustment are completed in step S105, the retired power battery system is put back into operation. The data acquisition module in the system will continue to monitor the operating data of each individual battery cell in real time, including voltage, current, temperature, internal resistance, charge / discharge capacity, and cycle count. This data is collected and transmitted to the data storage and management server at a high frequency, forming a continuously updated real-time operating database. The monitoring frequency is consistent with the data acquisition frequency in step S101, for example, 100 Hz.

[0071] Secondly, evaluate the accuracy of the degradation trend prediction model. Periodically, for example, every 24 hours or weekly, compare the actual monitored degradation of individual cells (e.g., actual capacity degradation percentage, actual internal resistance increment) with the degradation trend predicted by the model in step S103. Evaluate the model's prediction accuracy by calculating the prediction error, such as mean absolute error or mean squared error. If the model's prediction error consistently exceeds the preset allowable range, or if the model's recall rate for identifying high-risk degradation cells decreases, model retraining is required.

[0072] Subsequently, the effectiveness of the dynamic clustering grouping strategy is evaluated. The effect of the grouping strategy is assessed by analyzing the actual operating data of the optimized battery pack. The evaluation metrics include, but are not limited to, the voltage consistency, internal resistance consistency, and temperature consistency of individual cells within the battery pack, as well as the overall energy output efficiency and cycle life of the battery pack. For example, the consistency level is evaluated by calculating the standard deviation of the voltage or internal resistance of individual cells within the same battery pack. If the consistency index is lower than a preset threshold, or if a new "weakest link" phenomenon occurs, it indicates that the current clustering grouping strategy has room for improvement.

[0073] Furthermore, the system performs iterative updates to the model and strategy. When evaluation results show a decline in the performance of the prediction model or grouping strategy, the system automatically triggers the iterative update process. The iterative update includes: First, retrain the degradation trend prediction model. Update the training dataset from step S102 using the latest collected battery operating data and historical degradation data. Retrain the degradation trend prediction model on this updated dataset. This process can employ incremental learning or transfer learning methods to effectively utilize historical knowledge and adapt to the new data distribution, ensuring the model can capture the latest degradation characteristics and changes in operating conditions.

[0074] Second, parameter optimization of the dynamic clustering grouping strategy. Based on the evaluation results, adjust the parameters of the dynamic clustering algorithm in step S104, such as feature weights, the number of clusters K, the similarity metric, or update the threshold used for risk level assessment. These parameters can be optimized using intelligent optimization algorithms, such as genetic algorithms or particle swarm optimization algorithms, by searching and adjusting based on actual performance.

[0075] Finally, steps S103 to S105 are repeated. After the model and strategy are updated, the system will use the updated degradation trend prediction model to re-predict the future degradation state and risk level of all individual battery cells. Then, based on the new prediction results and the optimized grouping strategy, the retired power battery system is dynamically clustered again, and a new battery pack configuration optimization and adjustment scheme is generated. This closed-loop process continues, forming a self-learning and self-optimizing mechanism to ensure that the retired power battery system always operates in the optimal reliability state, while effectively reducing the frequency of grouping adjustments from twice a day to once a day, further reducing battery disassembly and assembly losses.

[0076] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "inclusion" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A reliability optimization design method for a retired power battery system, characterized in that, include: Obtain real-time operating data and historical degradation data of the power batteries to be restructured and retired; The real-time operating data and historical attenuation data are preprocessed. Based on the preprocessed historical degradation data, a single cell degradation trend prediction model is constructed and trained, and the model is used to predict the degradation trend of key performance parameters of each single cell in the future preset time period. Based on the real-time operating data, historical degradation data, and predicted degradation trends, a multi-dimensional consistency evaluation index system is constructed, and the comprehensive consistency evaluation score of each individual battery cell is calculated. Based on the comprehensive consistency evaluation score, the power batteries to be reorganized and retired are grouped using a dynamic clustering algorithm, so that the comprehensive consistency evaluation score of individual batteries in the same group reaches a preset threshold and the performance difference between groups is maximized. The dynamic grouping strategy is generated and executed, the recombined battery system operating performance data is monitored and recorded, and the operating performance data is fed back to the data acquisition step for iterative optimization.

2. The reliability optimization design method for retired power battery systems according to claim 1, characterized in that, The acquisition of real-time operating data and historical degradation data of the power batteries to be restructured and retired includes: Obtain real-time voltage data through the battery management system; Real-time current data is acquired using a Hall effect sensor. Real-time temperature data is acquired through an embedded thermistor array; Real-time internal resistance data were obtained using an AC impedance spectroscopy analyzer. Real-time state of charge data can be calculated and obtained using the Coulomb integral method or the open-circuit voltage method. Historical capacity decay data is obtained through periodic capacity testing, which employs constant current and constant voltage charging and constant current discharging modes. Accumulated cycle count data can be obtained through a battery management system or an external data logger.

3. The reliability optimization design method for retired power battery systems according to claim 2, characterized in that, The data preprocessing of the real-time operating data and historical attenuation data includes: The 3σ principle or the local outlier algorithm is used to detect outliers in the collected data, and median filtering or a statistical model based on historical data is used to correct outliers. Missing data points can be filled using linear interpolation or cubic spline interpolation methods; Dynamic time warping algorithms or synchronous sampling techniques are used to align the time series of data from different sensors, unifying all data to a preset sampling frequency. All feature data are normalized using the minimum-maximum normalization method or the Z-score normalization method.

4. The reliability optimization design method for retired power battery systems according to claim 3, characterized in that, The construction and training of a single-cell degradation trend prediction model, and the use of the model to predict the degradation trend of key performance parameters of each single cell within a future preset time period, includes: The preprocessed historical decay data is divided into a training dataset and a validation dataset. Construct a deep learning model based on a temporal convolutional network or a long short-term memory network. The input layer of the model receives multi-dimensional temporal features of a single battery cell, and the output layer outputs predicted values ​​of capacity decay rate and internal resistance growth rate within a preset time period in the future. The mean squared error is used as the loss function, and the deep learning model is trained using an adaptive moment estimation optimizer. Evaluate the model performance on the validation dataset until the preset prediction accuracy requirement is met.

5. The reliability optimization design method for retired power battery systems according to claim 4, characterized in that, The preset time period for the future is between 24 hours and 72 hours.

6. The reliability optimization design method for retired power battery systems according to claim 5, characterized in that, The construction of a multi-dimensional consistency evaluation index system and the calculation of the comprehensive consistency evaluation score for each individual battery cell include: The voltage consistency index is defined as the relative deviation of the current voltage of a single cell from the average voltage of all single cells in the set to be grouped, or the standard deviation of voltage within the set. The capacity consistency index is defined as the relative deviation between the current capacity of a single cell and the average capacity of all single cells in the set to be grouped, or the standard deviation of the set's capacity. The internal resistance consistency index is defined as the relative deviation between the current internal resistance of a single cell and the average internal resistance of all single cells in the set to be grouped, or the standard deviation of internal resistance within the set. The temperature consistency index is defined as the relative deviation of the current temperature of a single cell from the average temperature of all single cells in the set to be grouped, or the standard deviation of the temperature within the set. The consistency index of future degradation trend is defined as the relative deviation of the predicted capacity degradation rate of a single cell from the average predicted capacity degradation rate of the entire set or the standard deviation of the predicted capacity degradation rate of the entire set, and the relative deviation of the predicted internal resistance growth rate of a single cell from the average predicted internal resistance growth rate of the entire set or the standard deviation of the predicted internal resistance growth rate of the entire set. The weight coefficients of each consistency index are determined by the analytic hierarchy process (AHP) or principal component analysis (PCA), and the weighted sum is calculated as the overall consistency evaluation score.

7. The reliability optimization design method for retired power battery systems according to claim 6, characterized in that, The dynamic clustering algorithm is K-means++, DBSCAN, or spectral clustering. The clustering algorithm uses the comprehensive consistency evaluation score as the clustering feature for iterative calculation, so that the individual cells within the same cluster have high similarity, and the individual cells between different clusters have low similarity.

8. The reliability optimization design method for retired power battery systems according to claim 7, characterized in that, The generation and execution of the dynamic grouping strategy includes: Based on the clustering results of the dynamic clustering algorithm, a battery physical reorganization instruction is generated, which specifies the battery pack number to which each individual battery should belong; The instructions are transmitted to an automated battery reconfiguration device, which is responsible for disassembling, transporting, and reinstalling individual batteries to form a new battery pack. Functional tests were conducted on the recombined battery system to verify that its performance met the design requirements.

9. A reliability optimization design system for retired power battery systems, characterized in that, include: The data acquisition module is used to acquire real-time operating data and historical degradation data of the power batteries to be restructured and retired. The real-time operating data includes the voltage, current, temperature, internal resistance and state of charge of individual batteries. The historical degradation data includes the capacity, health status and cumulative number of cycles of each charge and discharge cycle of individual batteries. The data preprocessing module is used to perform data cleaning, missing value imputation, time series alignment, and feature normalization on the real-time running data and historical decay data. The degradation trend prediction module is used to construct and train a single cell degradation trend prediction model based on the preprocessed historical degradation data, and use the model to predict the degradation trend of key performance parameters of each single cell within a future preset time period. The key performance parameters include capacity degradation rate and internal resistance growth rate. The consistency evaluation module is used to construct a multi-dimensional consistency evaluation index system based on the real-time operating data, historical degradation data and predicted degradation trend, and calculate the comprehensive consistency evaluation score of each individual cell. The multi-dimensional consistency evaluation index system includes voltage consistency, capacity consistency, internal resistance consistency, temperature consistency and future degradation trend consistency. The dynamic grouping module is used to group the retired power batteries to be reorganized into groups based on the comprehensive consistency evaluation score using a dynamic clustering algorithm, so that the comprehensive consistency evaluation score of the individual batteries in the same group reaches a preset threshold and the performance difference between groups is maximized. The strategy execution and feedback module is used to generate and execute the dynamic grouping strategy, monitor and record the operating performance data of the recombined battery system, and feed the operating performance data back to the data acquisition module for iterative optimization.

10. The reliability optimization design system for retired power battery systems according to claim 9, characterized in that, The data acquisition module includes: Voltage sensors are used to monitor the voltage of individual battery cells in real time. Current sensor used to monitor the current of a single battery cell in real time; Temperature sensor used to monitor the temperature of individual battery cells in real time; Internal resistance testing unit, used to monitor the internal resistance of individual cells in real time; The state of charge calculation unit is used to calculate the state of charge of a single battery cell. The capacity testing equipment interface is used to connect to external capacity testing equipment to obtain historical capacity degradation data; The data storage unit is used to store all collected real-time operating data and historical attenuation data.

Images