Method for detecting, sorting, recombining and integrating waste power batteries

Abstract

The invention belongs to the technical field of waste power battery recovery, and discloses a method for detecting, sorting, recombining and integrating waste power batteries. The method comprises the following steps: acquiring state parameters of a waste power battery, and evaluating charge and health states through an evaluation model integrating model driving and data driving; based on an evaluation result, carrying out adaptive clustering preliminary sorting and feature extraction secondary grouping to obtain a consistency matching battery pack; in combination with the electrical parameter requirements of the optical storage charging station, a linear programming algorithm is adopted to determine series-parallel configuration and recombine; the active equalization battery management system monitors the operation state of the reconstitution battery system, and adjusts charging and discharging parameters in combination with scene working condition fluctuation. According to the invention, the consistency and the operation stability of the recombined waste power battery are improved, and the application requirements of a light storage charging station are met.

Description

Technical Field

[0001] This application relates to the field of waste power battery recycling technology, specifically to a method for detecting, sorting, and recombining waste power batteries. Background Technology

[0002] With the rapid development of the new energy industry, the recycling and utilization of spent power batteries has become a key focus of the industry. However, the current processing of spent power batteries still faces many technical challenges, restricting their resource utilization efficiency and application scope. Existing technologies often employ a single model-driven or data-driven approach to assess the state of spent power batteries, making it difficult to balance assessment accuracy and real-time performance. This results in insufficient accuracy in assessing the state of charge and health status. Battery sorting relies heavily on manual screening or simple clustering algorithms, leading to poor consistency among grouped batteries and failing to meet subsequent reconfiguration requirements. The battery reconfiguration process does not consider the electrical parameter requirements of specific application scenarios, relying solely on experience to determine series and parallel connection methods, resulting in low adaptability of the reconfigured battery system to the application scenario. Furthermore, the operation and management of the reconfigured battery system lacks linkage with the operating conditions of the scenario, making it difficult to adjust charging and discharging parameters according to fluctuations in operating conditions. This leads to insufficient operational stability, shortened lifespan, and an inability to efficiently adapt to the actual application needs of specific scenarios. Summary of the Invention

[0003] To solve, or at least partially solve, the above-mentioned technical problems, this application provides a method for the detection, sorting, recombining and integration of waste power batteries.

[0004] This application provides a method for the detection, sorting, and reassembly of waste power batteries, including the following steps: S1. Obtain the state parameters of the spent power battery and evaluate the state of charge and health of the spent power battery through an evaluation model that integrates model-driven and data-driven approaches to obtain the state evaluation result. S2. Based on the state assessment results, an adaptive clustering algorithm is used to perform preliminary sorting of the waste power batteries. Then, feature extraction and analysis are used to perform secondary grouping of the pre-sorted waste power batteries to obtain battery packs with consistent matching. S3. Based on the electrical parameter requirements of the photovoltaic-storage charging station scenario, a linear programming algorithm is used to determine the series and parallel configuration of the consistent matching battery packs, and the consistent matching battery packs are recombined to form a recombined battery system. S4. The operating status of the recombinant battery system is monitored in real time by the active equalization battery management system, and the charging and discharging parameters of the recombinant battery system are adjusted in combination with the operating condition fluctuations of the photovoltaic-storage charging station scenario.

[0005] Optionally, the evaluation model includes an unscented Kalman filter and a deep neural network, with the unscented Kalman filter providing model-driven support and the deep neural network providing data-driven support.

[0006] Optionally, the fusion method of the unscented Kalman filter and the deep neural network is as follows: The deep neural network is trained offline using historical cycle data of spent power batteries to obtain an initial state evaluation model. During online evaluation, the unscented Kalman filter is driven by the collected real-time state parameters to correct the output of the initial state evaluation model in real time, thereby obtaining the state evaluation result.

[0007] Optionally, the preliminary sorting of the spent power batteries using an adaptive clustering algorithm specifically includes the following steps: Extract the state parameters of the state of charge and the state parameters of the health state from the state assessment results, and calculate statistical characteristic values; The neighborhood radius of the density clustering algorithm is determined based on the statistical feature values. The neighborhood radius is positively correlated with the dispersion of the state parameters of the charged state and the state parameters of the healthy state. Based on the distribution density of the state parameters of the charged state and the state parameters of the healthy state, the minimum number of samples in the density clustering algorithm is adjusted. The higher the distribution density, the larger the value of the minimum number of samples. The density clustering algorithm is used to perform cluster analysis on waste power batteries, and battery packs with parameter differences that meet the requirements are selected to complete the preliminary sorting.

[0008] Optionally, the secondary grouping of the pre-sorted waste power batteries through feature extraction and analysis specifically includes the following steps: Extract the time series of state parameters of each waste power battery after preliminary sorting; The state parameter time series is transformed into a two-dimensional feature map using the Markov Transition Field, while preserving the correlation characteristics of the state parameters over time. The local window attention mechanism of the Swing Transformer is used to extract the local correlation features of the parameters in the two-dimensional feature map. Combined with the global feature fusion module, the overall distribution law of the parameters is determined to obtain the comprehensive feature vector of each waste power battery. Calculate the similarity between the comprehensive feature vectors of each waste power battery, and group the batteries whose similarity meets the preset conditions into the same group to complete the secondary grouping.

[0009] Optionally, after S2 and before S3, the following are also included: After secondary grouping and consistency matching, obtain the real-time voltage and capacity data of each waste power battery in the same group, and calculate the voltage deviation and capacity deviation between batteries in the group. Based on the aforementioned state assessment results and the operating condition data of the photovoltaic-storage charging station scenario, the target value for battery balance within the group is determined. Based on the voltage deviation, the capacity deviation, and the equalization target value, the batteries in the group are replenished or released; and during the replenishment or release process, the voltage and capacity changes of each battery in the group are monitored in real time, and the equalization rate is adjusted according to the changing trend until the voltage deviation and capacity deviation of the batteries in the group meet the recombination requirements, thus completing the equalization process.

[0010] Optionally, the step of using a linear programming algorithm to determine the series-parallel configuration of the consistent matching battery packs and reorganizing the consistent matching battery packs specifically includes the following steps: After equalization processing, obtain the voltage and internal resistance data of each waste power battery, as well as the nominal voltage and nominal capacity of the photovoltaic-storage charging station scenario. An integer linear programming model is constructed, with the number of series and parallel connections in the series-parallel configuration as decision variables, the nominal voltage and nominal capacity as constraints, and minimizing the voltage difference and internal resistance difference within the recombined battery pack as the optimization objective. Solve the integer linear programming model to obtain the optimal series-parallel configuration that satisfies the constraints and achieves the optimization objective; According to the optimal series-parallel configuration, the balanced waste power batteries are connected and assembled to form a recombinant battery system.

[0011] Optionally, the condition assessment process also includes prediction of the remaining life of the spent power batteries, and the prediction results are used for reconfiguration and adaptation, specifically including: Collect historical charge and discharge cycle data of used power batteries and operating condition data of photovoltaic-storage charging station scenarios; Based on the corrected state parameter values ​​of the state evaluation results from the unscented Kalman filter output, the deep neural network is optimized and trained to construct a remaining life prediction model. The prediction model incorporates the operating condition factors that affect battery life extracted from the operating condition data of the photovoltaic-storage charging station scenario. The remaining life prediction model is used to output the remaining life prediction results for each used power battery. The process of determining the series-parallel configuration and reconfiguration also includes: The remaining lifetime prediction results, along with voltage data and internal resistance data, are used as constraints to ensure that the difference in remaining lifetime among the batteries in the same recombinant battery system meets the preset lifetime requirements.

[0012] Optionally, the step of monitoring the operating status of the recombinant battery system in real time through the active balancing battery management system and adjusting the charging and discharging parameters of the recombinant battery system in combination with the operating condition data of the photovoltaic-storage charging station scenario specifically includes the following steps: The real-time voltage, real-time internal resistance, and real-time temperature of each individual cell in the recombinant battery system are obtained. Combined with the operating condition data of the photovoltaic-storage charging station scenario, the parameter thresholds and temperature protection thresholds for segmented charging are determined. The charging process employs a segmented charging mode of constant current-constant voltage-trickle charge. During the constant current stage, the current is used to charge the battery to meet the power requirements of the scenario. During the constant voltage stage, the real-time voltage is kept stable within the target voltage range. During the trickle charge stage, the capacity differences between individual battery cells are adjusted and compensated. During the charging process, the active equalization battery management system monitors the real-time temperature of each individual battery cell. When the real-time temperature reaches the temperature protection threshold, the charging rate is reduced. When the real-time temperature exceeds the temperature protection threshold, charging is paused and cooling measures are implemented. Charging is resumed after the real-time temperature drops below the temperature protection threshold.

[0013] Optionally, after the recombinant battery system completes active balancing management, it further includes the following steps: The output voltage and output power response characteristics of the recombinant battery system, as well as the grid connection voltage, frequency standard, and power fluctuations on the photovoltaic side of the power station are obtained. To meet the power grid access standards and reduce grid connection impact, a compatibility adaptation model is constructed. The output voltage of the recombinant battery system, the output power response characteristics, the access voltage of the power grid, the frequency standard, and the power fluctuation on the photovoltaic side are used as inputs to determine the power response adjustment threshold and voltage matching range. Based on the output results of the compatibility adaptation model, the output power response speed of the recombinant battery system is adjusted by the active equalization battery management system, and the power allocation ratio between the photovoltaic side and the battery side is adjusted to reduce the impact of power surges on the power grid.

[0014] The method provided in this application has the following beneficial effects: The method for detecting, sorting, and reassembling used power batteries provided in this application assesses the state of charge and health of used power batteries through a model-driven and data-driven evaluation model. This approach combines the advantages of both methods, improving the accuracy and reliability of the state assessment results and effectively avoiding misjudgments caused by the limitations of a single evaluation method, thus reducing the number of unqualified batteries entering subsequent processes. Based on the state assessment results, an adaptive clustering algorithm is first used for preliminary sorting, followed by secondary grouping through feature extraction and analysis. This enables refined screening of used power batteries, improves the consistency of the grouped battery packs, and solves the parameter dispersion problem caused by the wear and tear of used power batteries. According to the electrical parameter requirements of the photovoltaic-storage charging station scenario, a linear programming algorithm is used to determine the series and parallel configuration of the battery packs and complete the reassembly. This allows the resulting reassembled battery system to better meet the actual electrical needs of the photovoltaic-storage charging station, avoiding resource waste caused by blind reassembly and improving the reuse efficiency of used power batteries. By actively balancing the battery management system to monitor the operating status of the recombinant battery system in real time and adjusting the charging and discharging parameters according to the operating condition fluctuations in the photovoltaic-storage charging station scenario, the operating status of the recombinant battery system can be controlled in real time, adapting to changes in operating conditions in the scenario in a timely manner, ensuring the stable operation of the recombinant battery system in the photovoltaic-storage charging station scenario, extending the service life of the recombinant battery system, giving full play to the residual value of waste power batteries, reducing the operating cost of photovoltaic-storage charging stations, and providing a practical method for the tiered utilization of waste power batteries. Attached Figure Description

[0015] Figure 1 This application provides a schematic flowchart of a method for detecting, sorting, and reassembling waste power batteries, as shown in the embodiments of this application. Figure 2 This is a schematic diagram of a recombinant battery system provided in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0017] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0018] See Figures 1 to 2 This application provides a method for detecting, sorting, and reassembling waste power batteries, including the following steps: S1. Obtain the state parameters of the spent power batteries and evaluate the state of charge and health of the spent power batteries through an evaluation model that integrates model-driven and data-driven approaches to obtain the state evaluation results. S2. Based on the state assessment results, an adaptive clustering algorithm is used to perform preliminary sorting of waste power batteries. Then, through feature extraction and analysis, the waste power batteries after preliminary sorting are grouped again to obtain battery packs with consistent matching. S3. Based on the electrical parameter requirements of the photovoltaic-storage charging station scenario, a linear programming algorithm is used to determine the series and parallel configuration of the battery packs with consistent matching, and the battery packs with consistent matching are recombined to form a recombined battery system. S4. The active equalization battery management system monitors the operating status of the recombined battery system in real time and adjusts the charging and discharging parameters of the recombined battery system in combination with the operating condition fluctuations of the photovoltaic-storage charging station scenario.

[0019] Specifically, in response to the problems of insufficient accuracy in state assessment, poor consistency in sorting results, weak adaptability of the recombining system, and operation and maintenance not being in line with application scenarios in the process of cascade utilization of waste power batteries, this method proposes an integrated processing approach that carries out assessment, sorting, recombining, and operation parameter control in stages, so as to achieve efficient cascade utilization of waste power batteries and meet the actual application needs of photovoltaic-storage charging stations.

[0020] First, various state parameters of the spent power batteries are collected, such as voltage, internal resistance, charge-discharge cycle count, capacity decay rate, and temperature characteristics. These parameters comprehensively reflect the basic performance and wear and tear of the batteries. Based on this, an evaluation model integrating model-driven and data-driven approaches is used to comprehensively assess the battery's state of charge and health. This evaluation model combines the advantages of both approaches to more comprehensively reflect the actual performance of the batteries, ultimately yielding valuable state assessment results. Based on the obtained state assessment results, an adaptive clustering algorithm is first used to initially sort the spent power batteries. This algorithm categorizes the battery performance parameters, selecting groups of batteries with similar performance. Then, through feature extraction and analysis, the initially sorted batteries are regrouped to further refine the performance classification criteria, improve the performance consistency within battery packs, and ultimately obtain battery packs with consistent performance.

[0021] Based on the electrical parameter requirements of the photovoltaic-energy storage charging station scenario, a suitable linear programming algorithm is selected. Taking the performance matching degree of the battery pack and the electrical requirements of the scenario as the core considerations, a series-parallel configuration scheme for the battery pack with consistent matching is determined. The battery pack is then reassembled according to this scheme to form a reassembled battery system that can be directly applied to the photovoltaic-energy storage charging station. Figure 2 This is a schematic diagram of a recombinant battery system provided in an embodiment of this application. After the recombinant battery system is put into operation, an active balancing battery management system is used to monitor the system's operating status in real time, promptly grasp the operating status of each part of the system, and adjust the charging and discharging parameters of the recombinant battery system in combination with the operating condition fluctuations of the photovoltaic-storage charging station scenario, so that the system is always in an operating state adapted to the scenario requirements.

[0022] The above methods effectively improve the accuracy of state assessment of spent power batteries, avoiding sorting errors caused by assessment biases. The refined sorting and grouping process significantly improves the consistency of battery packs, laying a solid foundation for subsequent remanufacturing. The remanufacturing scheme based on linear programming algorithms allows the remanufactured battery system to better meet the electrical requirements of photovoltaic-energy storage charging stations, reducing resource waste and improving the reuse efficiency of spent power batteries. Real-time monitoring and dynamic parameter adjustment of the active balancing battery management system ensures stable operation of the remanufactured battery system under changing operating conditions, extending the system's lifespan, fully utilizing the residual value of spent power batteries, reducing the operating costs of photovoltaic-energy storage charging stations, and providing a feasible implementation path for the tiered utilization of spent power batteries, aligning with the development direction of resource recycling.

[0023] In some implementations, the evaluation model includes unscented Kalman filtering and deep neural networks, with unscented Kalman filtering providing model-driven support and deep neural networks providing data-driven support.

[0024] Specifically, the assessment model used to evaluate the state of charge and health of spent power batteries combines model-driven and data-driven approaches. It consists of two parts: an unscented Kalman filter and a deep neural network. The unscented Kalman filter supports the model-driven approach, leveraging its system state equation-based characteristics to extrapolate the changing trends of battery state parameters. This effectively captures the correlation characteristics of parameters during dynamic changes, compensating for the assessment bias of purely data-driven models lacking prior knowledge. The deep neural network supports the data-driven approach, utilizing its ability to fit massive amounts of data to analyze hidden parameter correlation patterns in the historical operating data of spent power batteries, thus improving the assessment model's adaptability to battery states under complex operating conditions.

[0025] The combination of the two technologies retains the rigor of the model-driven approach based on theoretical derivation while incorporating the flexibility of the data-driven approach based on actual data. This allows the evaluation model to have higher accuracy and stability when dealing with the status evaluation tasks of waste power batteries with different aging levels and different operating histories. This approach differs from the conventional single-driven evaluation mode and can further enhance the reliability of the evaluation results.

[0026] In some implementations, the fusion of unscented Kalman filtering and deep neural networks is as follows: The initial state evaluation model is obtained by offline training of a deep neural network using historical cycle data of spent power batteries. During online evaluation, the unscented Kalman filter is driven by the collected real-time state parameters to correct the output of the initial state evaluation model in real time, thus obtaining the state evaluation result.

[0027] Specifically, the process begins with an offline training phase, collecting a large amount of historical cycle data from spent power batteries. This data includes key parameters such as voltage, current, and temperature under different charge / discharge rates and usage durations. This historical cycle data is then divided into training and validation sets and input into a deep neural network for model training. By continuously adjusting the network's weights and bias parameters, the deep neural network learns the changing patterns of battery state parameters from the historical data, until the error between the initial state assessment model's output and the actual battery state stabilizes within an acceptable range, thus completing the construction of the initial state assessment model.

[0028] After entering the online evaluation phase, the battery management system collects real-time state parameters of the spent power batteries and synchronously inputs these parameters into a pre-built unscented Kalman filter and initial state evaluation model. The unscented Kalman filter, based on its own state and observation equations, analyzes and processes the real-time collected parameters. It then corrects the state of charge and health status assessment results output by the initial state evaluation model in real time, taking into account the dynamic changes in the real-time parameters. The correction process primarily addresses the evaluation bias caused by the initial model's failure to consider real-time operating condition fluctuations.

[0029] This fusion of offline training to build the initial model and online real-time correction not only utilizes historical data to ensure the basic evaluation capability of the initial state assessment model, but also improves the adaptability of the assessment model to dynamic working conditions through real-time parameter correction. Unlike the conventional method of directly using a fixed model for evaluation, this effectively improves the reliability of the state assessment results.

[0030] Considering the specific operating conditions of photovoltaic-energy storage charging stations, to further improve the adaptability of the fusion evaluation model, adaptive adjustments to the structure of unscented Kalman filters or deep neural networks can be considered. Taking the unscented Kalman filter as an example, its original structure uses a fixed sampling strategy and noise covariance matrix. When facing complex operating conditions such as frequent fluctuations in charging and discharging power and sudden temperature changes in photovoltaic-energy storage charging stations, it is prone to insufficient sampling accuracy and delayed response to parameter mutations. To address this issue, an adaptive sampling mechanism can be introduced in the sampling stage. The number and distribution density of sampling points are adjusted according to the real-time operating conditions of the photovoltaic-energy storage charging station. When operating conditions fluctuate drastically, the number of sampling points is increased and their distribution is densified to improve the ability to capture parameter mutations. Simultaneously, the originally fixed noise covariance matrix is ​​replaced with an updatable covariance matrix. By monitoring the changing trend of the evaluation error in real time and adjusting the matrix parameters, the unscented Kalman filter can better adapt to the fluctuations in operating conditions within the scenario. If we choose to improve the deep neural network structure, its original network structure often has fixed layers and a fixed number of neurons, which is insufficient for mining the temporal correlation features of battery state parameters in the photovoltaic-storage-charging station scenario. We can consider adding a temporal attention module to the network to strengthen the focus on battery charging and discharging time-series data. Simultaneously, based on the parameter characteristics under different operating conditions within the scenario, we can adjust the number of neurons in each layer of the network to improve the adaptability of the deep neural network to the scenario's operating conditions. Through the above solutions, the fused evaluation model can better fit the actual operating conditions of the photovoltaic-storage-charging station, further improving the accuracy and stability of the state assessment.

[0031] In some implementations, an adaptive clustering algorithm is used for preliminary sorting of used power batteries, specifically including the following steps: Extract the state parameters of the charged state and the state parameters of the healthy state from the state assessment results, and calculate the statistical characteristic values; The neighborhood radius of the density clustering algorithm is determined based on statistical characteristic values. The neighborhood radius is positively correlated with the dispersion of the state parameters of the charged state and the state parameters of the healthy state. Based on the distribution density of state parameters of the charged state and state parameters of the healthy state, the minimum number of samples in the density clustering algorithm is adjusted. The higher the distribution density, the larger the minimum number of samples. Density clustering algorithm was used to perform cluster analysis on waste power batteries, and battery packs with parameter differences that met the requirements were selected to complete the preliminary sorting.

[0032] Specifically, the process begins with extracting state parameters and calculating statistical features. Core state parameters for the state of charge (SOC) and health state are extracted from the state assessment results. The core SOC parameters include the remaining percentage of charge, open-circuit voltage, and charge / discharge efficiency. Core health state parameters include the capacity decay rate, internal resistance growth rate, and remaining cycle life. These state parameters directly reflect the battery's remaining capacity and degree of degradation. Statistical analysis methods are then used to calculate the dispersion, mean, and variance of these state parameters. The dispersion value serves as a core reference for subsequently adjusting the clustering algorithm parameters, while the mean and variance are used to help determine the overall distribution of the parameters.

[0033] Next, we proceed to the parameter adjustment stage of the clustering algorithm. First, we determine the neighborhood radius of the density clustering algorithm. Conventional density clustering algorithms often use a fixed neighborhood radius, which is difficult to adapt to situations where the parameter dispersion varies greatly among different batches of spent power batteries. In this embodiment, the neighborhood radius is positively correlated with the dispersion of the state of charge (SCC) and state of health (SOH) parameters. That is, when the parameter dispersion is large, the neighborhood radius is appropriately increased to avoid over-splitting batteries with similar performance; when the parameter dispersion is small, the neighborhood radius is reduced to improve the precision of sorting. Subsequently, we adjust the minimum sample size of the density clustering algorithm. By analyzing the distribution density of the SCC and SOH parameters, the region with higher distribution density has a larger minimum sample size to ensure that the battery packs selected in that region have a sufficient quantity to meet the batch requirements of subsequent reassembly. Conversely, the region with lower distribution density has a lower minimum sample size to avoid wasting effective battery resources due to excessively high sample size requirements.

[0034] Finally, cluster analysis and screening were conducted. The adjusted density clustering algorithm was applied to analyze the state of charge and health parameters of all spent power batteries. The density clustering algorithm automatically identifies regions with high parameter distribution density, grouping batteries within the same high-density region with small parameter differences into the same group, and removing batteries with abnormal parameters or those that cannot be classified into any high-density region. This initial sorting process yields battery packs with basic consistency, laying the foundation for further consistency improvement in subsequent secondary grouping. This parameter-adaptive clustering method, compared to fixed-parameter clustering algorithms, is better suited to the highly discrete nature of spent power battery parameters, improving the rationality and accuracy of the sorting.

[0035] In some implementations, the pre-sorted waste power batteries are regrouped through feature extraction and analysis, specifically including the following steps: Extract the time series of state parameters of each waste power battery after preliminary sorting; The state parameter time series is transformed into a two-dimensional feature map by using the Markov Transition Field, while preserving the correlation characteristics of the state parameters over time. The local window attention mechanism of the Swing Transformer is used to extract the local correlation features of the parameters in the two-dimensional feature map. Combined with the global feature fusion module, the overall distribution law of the parameters is determined, and the comprehensive feature vector of each waste power battery is obtained. Calculate the similarity between the comprehensive feature vectors of each waste power battery, and group the batteries whose similarity meets the preset conditions into the same group to complete the secondary grouping.

[0036] After the initial sorting is completed, a second grouping is required through feature extraction and analysis to further improve the consistency of the battery packs. The specific implementation process is roughly as follows: extraction of state parameter time series, transformation of two-dimensional feature map, extraction of comprehensive feature vector, and similarity matching grouping.

[0037] Specifically, the first step is to extract the state parameter time series, collecting continuous state parameter data of each waste power battery in each group after preliminary sorting at different charge-discharge cycle stages, forming a state parameter time series. Here, the state parameter time series refers to a data sequence formed by arranging the continuously collected state of charge and health-related parameters of the battery in chronological order over a period of time. This sequence reflects the changing pattern of battery performance over time and, compared to parameters at a single point in time, more comprehensively reflects the actual performance differences of the batteries.

[0038] Next, the state parameter time series is transformed into a two-dimensional feature map using the Markov Transition Field method. This method maps one-dimensional time series data into a two-dimensional matrix by calculating the transition probabilities between different parameter values ​​in the time series, thus forming a two-dimensional feature map. Its core advantage is that it completely preserves the correlation characteristics of state parameters over time during the transformation process, avoiding the loss of temporal correlation information in traditional feature extraction methods. During the transformation, the state parameter time series is first standardized to eliminate the influence of differences in parameter magnitudes. Then, reasonable interval partitioning rules are set to divide the parameter values ​​into several intervals. The transition probabilities of parameters between different intervals at adjacent time steps are calculated, ultimately constructing a two-dimensional feature map that intuitively reflects the temporal correlation characteristics of the parameters.

[0039] Subsequently, a comprehensive feature vector extraction is performed using the Swin Transformer model. The core structure of the Swin Transformer is a local window attention mechanism, which, compared to the global attention mechanism of traditional Transformers, can more efficiently capture local feature details. In implementation, the 2D feature map is first divided into multiple non-overlapping local windows. The association weights of pixels within each window are calculated using the local window attention mechanism to extract the local association features of the parameters. Then, a global feature fusion module fuses the features from each local window to analyze the overall distribution pattern of the parameters. During the extraction process, the level of feature abstraction can be gradually improved by stacking multiple layers of networks, ultimately outputting a comprehensive feature vector that fully characterizes battery performance.

[0040] Finally, similarity matching and grouping are performed. The cosine similarity between the comprehensive feature vectors of each battery within the same initial sorting group is calculated. Cosine similarity quantifies the degree of similarity between two vectors; the closer the value is to 1, the more similar the performance characteristics of the two batteries. Batteries whose cosine similarity meets the preset conditions are grouped together, completing the secondary grouping. This grouping method, based on the fusion of temporal features and deep semantic features, can more accurately identify battery performance differences and significantly improve the consistency of the grouped battery packs compared to traditional single-parameter-based grouping methods.

[0041] Furthermore, considering the frequent fluctuations in the operating conditions and varying degrees of performance degradation of spent power batteries in the photovoltaic-storage charging station scenario, adaptive structural adjustments to the Swin Transformer model can be considered to further improve the accuracy and efficiency of comprehensive feature vector extraction. The original Swin Transformer model uses a fixed-size local window partitioning method, and the attention calculation range within the window is fixed. When processing the two-dimensional feature map formed by battery state parameters in the photovoltaic-storage charging station scenario, it may have difficulty accurately adapting to the differences in parameter feature distribution of batteries with different degrees of degradation, and its ability to capture key local features is limited. To address this issue, two structural adjustments were made: First, the fixed window partitioning was replaced with dynamic window partitioning. The window size and number were adjusted based on the density distribution of parametric features in the 2D feature map. In regions with dense and rapidly changing parametric features, the window size was reduced to improve local feature capture accuracy, while in regions with smooth parametric features, the window size was increased to improve computational efficiency. Second, a condition-adaptive weight module was added to the local window attention mechanism. This module, combined with common operating condition data such as charging and discharging power fluctuations and temperature changes in photovoltaic-storage charging stations, assigned differentiated weights to features in different window regions. This strengthened the weights of features strongly correlated with operating condition changes and weakened the influence of irrelevant interference features. Through these adjustments, the Swin Transformer model can better reflect the actual operating conditions of photovoltaic-storage charging stations, more accurately extract the core features reflecting battery performance in this scenario, and thus make the secondary grouping results more reliable, improving the consistency of battery performance within the same group in the actual operating environment of the photovoltaic-storage charging station.

[0042] In some implementations, after S2 and before S3, the following steps are also included: After secondary grouping and consistency matching, obtain the real-time voltage and capacity data of each waste power battery in the same group, and calculate the voltage deviation and capacity deviation between batteries in the group. Based on the condition assessment results and the operating data of the photovoltaic-storage charging station scenario, the target value for battery balance within the group is determined. Based on voltage deviation, capacity deviation, and equalization target value, the batteries in the group are replenished or released; and during the replenishment or release process, the voltage and capacity changes of each battery in the group are monitored in real time, and the equalization rate is adjusted according to the changing trend until the voltage deviation and capacity deviation of the batteries in the group meet the recombination requirements, thus completing the equalization process.

[0043] After completing the secondary grouping to obtain a consistent matching battery pack, targeted equalization processing can be performed before determining the series and parallel configuration for reorganization, thereby further reducing the parameter differences of the batteries within the group and meeting the reorganization application requirements of the photovoltaic-storage charging station scenario.

[0044] Specifically, the process begins by collecting data on the acquisition and calculation deviations. This involves real-time acquisition of voltage and capacity data for each spent power battery within the same group after secondary grouping. Voltage data reflects the battery's instantaneous energy storage state, while capacity data reflects its maximum energy storage capacity. Based on the collected data, the voltage and capacity deviations between any two batteries within the group are calculated. By statistically analyzing the average voltage and capacity deviations of all batteries within the group, the consistency level of the current battery pack can be intuitively determined.

[0045] Subsequently, the balancing target value was determined. Combining the previously obtained state assessment results and the operating condition data of the photovoltaic-storage charging station scenario, the balancing target value of the batteries within the group was clarified. The state assessment results include key information such as the battery's state of charge and health status, which can serve as the basis for determining the balancing target value. The operating condition data of the photovoltaic-storage charging station includes the conventional charge-discharge rate and operating temperature range within the scenario. These operating conditions affect the actual operating performance of the batteries. Therefore, when determining the balancing target value, it is necessary to consider the battery's operating requirements in this scenario so that the balanced battery pack can adapt to the scenario's operating conditions.

[0046] Then, based on the calculated voltage deviation, capacity deviation, and determined balancing target value, a balancing strategy is formulated. For batteries with low voltage and insufficient capacity, a small current is used to gradually increase their voltage and capacity; for batteries with high voltage and excess capacity, resistive discharge is used to slowly release their excess charge. During the replenishment or release process, the voltage and capacity change trends of each battery in the group can be continuously monitored. If it is found that the voltage or capacity of a certain battery is close to the corresponding balancing target value, the balancing rate is adjusted, and the replenishment or release current intensity is reduced to avoid irreversible damage to the battery due to overcharging or over-discharging.

[0047] The entire equalization process continues until the voltage and capacity deviations of all batteries in the group meet the preset recombination requirements. This dynamic adjustment of the equalization rate, unlike the traditional fixed-rate equalization method, can efficiently improve the consistency of batteries in the group while ensuring battery safety. It solves the problems of over-equalization or under-equalization that are prone to occur in traditional equalization methods, and provides battery packs with stable performance and consistent parameters for recombination.

[0048] In some implementations, a linear programming algorithm is used to determine the series-parallel configuration of the battery packs with consistent matching, and the consistent matching battery packs are reorganized, specifically including the following steps: After equalization processing, obtain the voltage and internal resistance data of each waste power battery, as well as the nominal voltage and nominal capacity of the photovoltaic-storage charging station scenario. An integer linear programming model is constructed, with the number of series and parallel connections in the series-parallel configuration as decision variables, the nominal voltage and nominal capacity as constraints, and the optimization objective as minimizing the voltage difference and internal resistance difference within the recombined battery pack. Solve the integer linear programming model to obtain the optimal series-parallel configuration that satisfies the constraints and achieves the optimization objective; According to the optimal series and parallel configuration, the waste power batteries after equalization are connected and assembled to form a recombinant battery system.

[0049] After the equalization process is completed, the series and parallel configurations are determined and recombined. The core is to use a linear programming algorithm to match the battery pack configuration with the electrical requirements of the photovoltaic-storage charging station scenario, so that the resulting recombined battery system has stable performance and is adapted to the scenario requirements.

[0050] First, relevant parameters were collected, including comprehensive data on the voltage and internal resistance of each spent power battery after equalization processing. Simultaneously, the nominal voltage and nominal capacity requirements for the photovoltaic-energy storage charging station scenario were defined. The nominal voltage is the standard voltage value required for normal operation of the equipment under the specified scenario, and the nominal capacity is the minimum energy storage capacity required for the equipment to operate.

[0051] Next, an integer linear programming model is constructed. First, decision variables are defined, with the number of series and parallel connections in the series-parallel configuration set as positive integers, representing the number of batteries connected in series and the number of parallel branches in the battery pack, respectively. Then, constraints are set. According to circuit principles, the product of the number of series connections and the voltage of a single battery must meet the nominal voltage requirement, and the product of the number of parallel connections and the capacity of a single battery must meet the nominal capacity requirement. Based on this, equality or inequality constraints are constructed to ensure the configuration results meet the minimum electrical parameters of the scenario. Finally, the optimization objective is determined, with minimizing the voltage and internal resistance differences within the reconfigured battery pack as the core objective. Excessive differences in voltage and internal resistance within the pack can lead to uneven load distribution among the batteries during charging and discharging, accelerating the degradation of some batteries and affecting the overall system lifespan. By constructing the objective function, the quantitative indicators of voltage and internal resistance differences are incorporated into the optimization scope of the integer linear programming model.

[0052] The model is then solved using an algorithm adapted for integer programming problems. The collected parameter data and the constructed model parameters are input, and the algorithm iteratively calculates the optimal combination of series and parallel connections that satisfies all constraints and optimizes the objective. During the solution process, the feasibility of the solution needs to be verified. If multiple feasible solutions exist, the one with the smaller difference in voltage and internal resistance is prioritized.

[0053] Finally, the reassembly process is implemented. Following the optimal series-parallel configuration scheme, the balanced spent power batteries are connected and assembled to form a reassembled battery system. This configuration method, based on integer linear programming, compared to traditional empirical series-parallel configuration methods, can balance scenario requirements and battery consistency, improve the operational stability of the reassembled battery system, and maximize the utilization of spent power battery resources, avoiding resource waste or system failures caused by unreasonable configuration.

[0054] In some implementations, the condition assessment process also includes prediction of the remaining life of the spent power batteries, and the prediction results are used for reconfiguration and adaptation, specifically including: Collect historical charge and discharge cycle data of used power batteries and operating condition data of photovoltaic-storage charging station scenarios; Based on the corrected state parameter values ​​of the state assessment results from the unscented Kalman filter output, the deep neural network is optimized and trained to construct a remaining life prediction model. The prediction model incorporates the operating condition factors that affect battery life extracted from the operating condition data of the photovoltaic-storage charging station scenario. The remaining life prediction model is used to output the remaining life prediction results of each waste power battery. The process of determining series and parallel configurations and reconfiguration also includes: The remaining lifetime prediction results, along with voltage and internal resistance data, are used as constraints to ensure that the differences in remaining lifetime among the batteries in the same recombinant battery system meet the preset lifetime requirements.

[0055] While assessing the state of charge and health of spent power batteries, the remaining life can be predicted simultaneously, and the prediction results can be integrated into the series-parallel configuration and reconfiguration process to further improve the long-term operational stability of the reconfigured battery system.

[0056] First, data collection is conducted. In addition to collecting the state parameters required for evaluation, historical charge-discharge cycle data of spent power batteries and operating condition data of photovoltaic-storage charging station scenarios also need to be collected. Historical charge-discharge cycle data includes information such as the starting voltage, ending voltage, charge-discharge current, and cycle duration of each previous charge-discharge cycle, which can reflect the degradation pattern of batteries after long-term use. Operating condition data of photovoltaic-storage charging station scenarios includes parameters such as the conventional charge-discharge rate, average daily charge-discharge cycles, operating temperature fluctuation range, and peak power demand. These parameters all affect the battery degradation rate and are operating condition influencing factors that affect battery life.

[0057] Next, a remaining battery life prediction model was constructed, which was optimized based on the deep neural network in the original fusion assessment model. Specifically, the deep neural network was trained a second time using the corrected state parameter values ​​from the unscented Kalman filter output. These corrected state parameter values ​​are the state of charge and health parameters corrected in real time by the unscented Kalman filter, which are more accurate than the original assessment parameters. During training, historical charge-discharge cycle data was used as the basic training data, and operating condition influencing factors extracted from the photovoltaic-storage charging station operating condition data were incorporated into the model as feature inputs. This allows the model to learn the battery life degradation patterns under different operating conditions, avoiding prediction biases caused by traditional prediction models that do not consider scenario differences. After training, the model's prediction accuracy was verified using validation set data until the error between the model's output remaining battery life prediction result and the actual remaining battery life met the preset requirements, thus completing the construction of the remaining battery life prediction model.

[0058] Then, the remaining life prediction is performed. The state parameters of the spent power batteries collected in real time are input into the constructed remaining life prediction model. The model combines the influencing factors of the scenario and outputs the remaining life prediction results of each battery. The results reflect the estimated time that the battery can still operate stably in the photovoltaic-storage charging station scenario.

[0059] In determining the series-parallel configuration and reconfiguration process, in addition to the existing voltage and internal resistance data constraints, a new constraint condition is added: the predicted remaining lifespan. Specifically, the difference in remaining lifespan among the batteries in the same group is calculated, ensuring that the difference meets the preset lifespan requirement. Batteries with similar remaining lifespans are then grouped into the same reconfigured battery system. This reconfiguration method, which incorporates remaining lifespan constraints, differs from the traditional configuration method that only considers voltage and internal resistance. It avoids the problem of some batteries failing prematurely due to excessive lifespan differences, thus affecting the operation of the entire system and significantly improving the long-term reliability and service life of the reconfigured battery system.

[0060] In some implementations, the active balancing battery management system monitors the operating status of the recombined battery system in real time and adjusts the charging and discharging parameters of the recombined battery system based on the operating data of the photovoltaic-storage charging station scenario. Specifically, this includes the following steps: The real-time voltage, real-time internal resistance and real-time temperature of each individual cell in the recombinant battery system are obtained. Combined with the operating condition data of the photovoltaic-storage charging station scenario, the parameter thresholds and temperature protection thresholds for segmented charging are determined. The charging process employs a segmented charging mode of constant current-constant voltage-trickle charge. During the constant current stage, the current is used to charge the battery to meet the power requirements of the scenario. During the constant voltage stage, the real-time voltage is kept stable within the target voltage range. During the trickle charge stage, the current is adjusted to compensate for the capacity differences between individual battery cells. During charging, the active equalization battery management system monitors the real-time temperature of each individual battery cell. When the real-time temperature reaches the temperature protection threshold, the charging rate is reduced. When the real-time temperature exceeds the temperature protection threshold, charging is paused and cooling measures are implemented. Charging is resumed after the real-time temperature drops below the temperature protection threshold.

[0061] After the recombined battery system is assembled and put into operation in the photovoltaic-storage charging station scenario, it is necessary to perform real-time monitoring and parameter adjustment based on the active balancing battery management system to ensure the stability and safety of the system operation.

[0062] First, a monitoring module is deployed, with voltage sensors, current sensors, and temperature sensors installed at each individual battery cell and the total output terminal of the battery pack. The voltage sensors collect the real-time terminal voltage of the individual cells and the total output voltage of the battery pack; the current sensors collect the real-time charging and discharging current of the battery pack; and the temperature sensors collect the surface temperature of the individual cells and the ambient temperature of the battery pack.

[0063] Subsequently, relevant data is collected. The main control unit of the active balancing battery management system continuously receives real-time voltage, current, and temperature data from various sensors according to a preset acquisition frequency. The acquisition frequency can be set according to the operating characteristics of the photovoltaic-storage charging station. During stable operating conditions, the acquisition frequency can be appropriately reduced to save system energy consumption, while during periods of drastic fluctuations in operating conditions, the acquisition frequency can be increased to promptly capture changes in parameters.

[0064] Next, the operating status is analyzed and judged. Based on the pre-processed monitoring data, the main control unit calculates key indicators such as the voltage difference between individual battery cells, the charging and discharging power of the battery pack, and the temperature change rate of individual battery cells. The voltage difference refers to the absolute voltage difference between individual cells within the same battery pack, reflecting the pack's balance. The charging and discharging power is the output or input power of the battery pack calculated based on voltage and current data; this indicator must match the load demand of the photovoltaic-storage charging station. The temperature change rate refers to the magnitude of temperature change of individual battery cells per unit time; this indicator can be used to determine whether the battery is in an abnormal heating state. The main control unit compares the calculated indicators with preset safety thresholds to determine whether the battery system is in normal operating condition.

[0065] Finally, the charging and discharging parameters are adjusted. When monitoring data shows that the charging and discharging power of the battery pack does not match the load requirements of the scenario, the main control unit promptly adjusts the magnitude of the charging and discharging current to enable the output or input power of the battery pack to quickly adapt to load changes. When monitoring data shows that the voltage difference of some battery cells exceeds the preset threshold, the main control unit activates the active balancing function, transferring energy from battery cells with higher voltage to those with lower voltage through energy transfer, thus reducing the voltage difference between cells. When monitoring data shows that the temperature of a battery cell exceeds the safety threshold, the main control unit reduces the charging and discharging current intensity and activates the heat dissipation device to prevent the battery from experiencing thermal runaway due to high temperature. This adjustment method based on real-time monitoring data, unlike the traditional fixed parameter management mode, ensures that the recombinant battery system is always in optimal operating condition, effectively extending the battery system's lifespan and reducing the operation and maintenance costs of the photovoltaic-storage charging station.

[0066] In some implementations, after the battery recombination system completes active balancing management, the following steps are also included: To obtain the output voltage and output power response characteristics of the recombinant battery system, as well as the grid connection voltage, frequency standard, and power fluctuations on the photovoltaic side of the power station; To meet the power grid access standards and reduce grid connection impact, a compatibility adaptation model is constructed. The output voltage and output power response characteristics of the recombinant battery system, the access voltage of the power grid, the frequency standard, and the power fluctuations on the photovoltaic side are used as inputs to determine the power response adjustment threshold and voltage matching range. Based on the output results of the compatibility adaptation model, the output power response speed of the recombinant battery system is adjusted by the active equalization battery management system, and the power distribution ratio between the photovoltaic side and the battery side is adjusted to reduce the impact of power fluctuations on the power grid.

[0067] After the battery reconfiguration system completes active balancing management and ensures that the consistency of batteries within the group meets the standards, in order to ensure its stable connection with the power grid of the photovoltaic-storage charging station, further grid compatibility adaptation adjustments can be implemented. The core is to reduce grid connection impact by matching grid standards and smoothing out power fluctuations.

[0068] First, multi-dimensional parameters are collected and calibrated. On one hand, the monitoring module of the active equalization battery management system continuously collects the output voltage and output power response characteristics of the battery system. The output power response characteristics refer to the response speed, fluctuation amplitude, and stabilization time of the battery system after receiving a power adjustment command, which affects the power matching effect during grid connection. On the other hand, the grid monitoring unit of the photovoltaic-storage charging station collects the grid access voltage standard and frequency standard of the power station. At the same time, the photovoltaic side monitoring equipment collects real-time power fluctuation data of the photovoltaic modules. Photovoltaic side power fluctuation refers to the instantaneous fluctuation of photovoltaic output power caused by changes in light intensity and ambient temperature, which is one of the main factors causing grid connection impact.

[0069] Next, a grid compatibility adaptation model is constructed. With the dual core objectives of meeting power station grid access standards and reducing grid connection impact, a multi-input, single-output adaptation model is built. The model input layer is defined by four core parameters: the output voltage and output power response characteristics of the reconfigurable battery system, the grid access voltage / frequency standard of the power station, and photovoltaic power fluctuations. The core calculation layer employs a multi-objective optimization algorithm, combining circuit principles and grid connection specifications to establish the correlation mapping relationship between parameters. For example, the voltage regulation requirement is calculated by the difference between the output voltage and the grid access voltage, and the power response compensation capability of the battery system is matched by the photovoltaic power fluctuation amplitude. Considering the randomness of photovoltaic power fluctuations and the dynamic changes in grid load at photovoltaic-storage charging stations, a real-time operating condition adaptive module can be integrated into the model. This allows the model to adjust parameter weights according to changes in grid load and the intensity of photovoltaic power fluctuations, avoiding the insufficient adaptability of traditional fixed models and improving the model's adaptability to various operating conditions.

[0070] Subsequently, key control parameters are determined through model solving. The calibrated parameters are input into the constructed compatibility adaptation model, which iteratively calculates and outputs two core results: first, the power response adjustment threshold, which is the maximum allowable fluctuation range of the recombinant battery system's output power and the upper limit of its response speed; exceeding this threshold will lead to excessive grid-connected power surges; second, the voltage matching range, which is the specific range within which the output voltage of the recombinant battery system needs to be maintained to match the grid connection voltage standard. For example, if the grid connection voltage standard is 380V±10%, the voltage matching range output by the compatibility adaptation model can fall within the range of 342V~418V. If the photovoltaic power fluctuation is large, such as an instantaneous fluctuation exceeding 20%, the compatibility adaptation model will correspondingly lower the power response adjustment threshold and increase the battery system's power response speed to smooth out the fluctuations.

[0071] Finally, dynamic adjustment and grid connection adaptation are implemented. Based on the power response adjustment threshold and voltage matching range output by the compatibility adaptation model, the active balancing battery management system performs the following steps: First, it adjusts the output power response speed of the reconfigured battery system. By optimizing the power regulation algorithm of the battery management system, when photovoltaic power fluctuations are small, the response speed is appropriately reduced to decrease internal battery energy consumption; when photovoltaic power fluctuations are severe, the response speed is increased to within the power response adjustment threshold to quickly compensate for photovoltaic power shortfalls or absorb excess power. Second, it adjusts the power distribution ratio between the photovoltaic side and the battery side. Through linkage control with the photovoltaic inverter, when excessive photovoltaic output power leads to grid connection impact risks, the energy storage absorption ratio on the battery side is increased; when photovoltaic output power is insufficient, the discharge output ratio on the battery side is increased, ensuring a stable connection of the total output power of the photovoltaic and battery sides to the grid. During the adjustment process, the monitoring unit of the photovoltaic-storage charging station is continuously linked to collect real-time data such as photovoltaic power fluctuations, grid load changes, and remaining battery system capacity. The weights of each input parameter are adjusted through the real-time adaptive module built into the compatibility adaptation model, and the power response adjustment threshold and voltage matching range are corrected in real time. At the same time, the grid connection voltage, frequency and power fluctuations are continuously monitored. If the parameters are found to exceed the corrected threshold range, they are adjusted again until all parameters meet the grid access standards.

[0072] This grid connection adjustment method based on a compatibility adaptation model differs from the traditional passive mode of connecting to the grid first and then making corrections. By matching grid standards in advance and actively smoothing power fluctuations, it reduces the impact of grid connection from the source. At the same time, combined with the ability to adjust under real-time operating conditions, it improves the compatibility between the recombinant battery system and the grid, ensuring the stable operation of the overall power supply system of the photovoltaic-storage charging station.

[0073] Furthermore, considering the extreme operating conditions of photovoltaic-storage charging stations, such as short-term high-power charging and discharging and sudden power outages and restarts, the accuracy of the state assessment model based solely on external operating parameters may fluctuate. Therefore, further optimization can be implemented to enhance adaptability. Specifically, an impedance testing module can be added during the state parameter acquisition stage to collect electrochemical impedance spectroscopy data from spent power batteries and extract core features such as the real part, imaginary part, and characteristic frequency of the impedance. These features directly reflect the aging state of the battery's internal electrode-electrolyte interface and are more fundamental attenuation characterization parameters. This data can then be fused with existing state of charge, health state parameters, and operating condition data to construct a multi-source feature input matrix. Simultaneously, a dual-channel attention fusion structure replaces the traditional feature splicing method. One channel handles static, essential features like electrochemical impedance spectroscopy, while the other handles external dynamic operating features. An attention mechanism dynamically allocates weights for the two types of features; for example, increasing the weight of dynamic operating features in the early stages of battery aging and increasing the weight of static, essential features in the later stages of aging. This allows for real-time adjustment of feature fusion, significantly improving the assessment accuracy under extreme operating conditions.

[0074] For example, considering that existing sorting and grouping schemes focus on short-term battery parameter consistency without taking into account long-term cycle life matching, there is a possibility that recombined battery packs may fail prematurely due to life differences during long-term operation. Therefore, the sorting and grouping strategy can be further optimized. Specifically, a cycle life decay rate model under different operating conditions can be established first through accelerated aging tests. This rate refers to the percentage of capacity decay per charge-discharge cycle, accurately reflecting long-term durability. Based on this, the predicted remaining cycle count for each battery under standard operating conditions in a photovoltaic-storage charging station can be calculated. Then, a multi-objective clustering optimization function can be constructed that considers minimizing parameter differences, minimizing cycle life differences, and adapting the grouping number to the recombining requirements. A hierarchical iterative clustering approach is adopted, first coarsely clustering within ±5% of the remaining cycle count difference, and then finely clustering within the larger categories based on the original parameters and time-series characteristics. This ensures that the grouped batteries meet both short-term operational consistency and long-term cycle life matching, improving the long-term operational stability of the recombined battery system.

[0075] Furthermore, considering that current solutions do not differentiate between battery aging stages, a uniform approach may damage severely aged batteries or underutilize the performance of mildly aged batteries. Therefore, an optimized solution of balance assurance and recombination assurance can be developed. Specifically, an aging stage identification model can be constructed based on health status values, electrochemical impedance characteristics, and cycle life residual values, classifying batteries into three categories: mild (SOH≥80%), moderate (60%≤SOH<80%), and severe (SOH<60%). The safe balancing current threshold and charge / discharge rate upper limit for each stage can be clearly defined. Then, the balancing mode can be adjusted accordingly: rapid balancing to improve efficiency for mildly aged batteries, stable balancing to balance efficiency and safety for moderately aged batteries, and slow balancing with low current to avoid damage for severely aged batteries. When configuring batteries in series and parallel, batteries in the same aging stage should be prioritized. When resources are insufficient and mixing is necessary, linear programming can be used to optimize battery placement and match corresponding balancing strategies, achieving personalized adaptation of balancing and recombination, thereby improving resource utilization and operational safety.

[0076] For example, considering that without further extending to the scenario of the coupling effect of uneven battery thermal distribution and electrical performance degradation in photovoltaic-storage charging stations, it may be difficult to predict potential faults such as thermal runaway and internal short circuits in advance, a predictive maintenance model that ensures thermal protection and electrical protection can be constructed. Specifically, distributed temperature sensors can be added inside the battery pack to collect internal temperature distribution data. Combined with voltage and current data, the charging and discharging efficiency and internal resistance heat generation power of each battery can be calculated. A thermal-guaranteed electrical coupling model can be constructed based on the heat conduction equation and circuit equation to quantitatively analyze the positive feedback loop relationship where increased internal resistance leads to increased heat generation and local high temperature accelerates internal resistance decay. A long short-term memory network is introduced to build a fault prediction model. Historical thermal-guaranteed electrical coupling monitoring data, state assessment results, and remaining life prediction values ​​are input to learn the parameter change patterns before the fault occurs. The model outputs a fault warning signal and locates the faulty battery 3 to 5 operating cycles in advance. When the warning signal is triggered, a warning-level operation and maintenance mode can be activated to reduce the charging and discharging rate of the branch where the faulty battery is located, enhance regional heat dissipation efficiency, optimize the equalization strategy to ensure voltage stability, and link with the photovoltaic and energy storage charging station energy management system to coordinate the charging and discharging rhythm of photovoltaic and energy storage to avoid the faulty battery bearing additional load.

[0077] For example, current solutions focus on the operation of individual recombinant battery systems, which means they do not consider scenarios that coordinate with the photovoltaic output fluctuations of photovoltaic charging stations and changes in grid load. This may lead to problems such as insufficient photovoltaic absorption and untimely grid replenishment during peak periods. To address this, a power regulation optimization scheme that integrates photovoltaic and energy storage could be developed. Specifically, a real-time communication interface can be set up between the active balancing battery management system and the photovoltaic-storage charging station energy management system to obtain scenario energy information such as real-time and predicted photovoltaic output data, grid load data, and time-of-use electricity price data. A multi-objective optimization model is constructed with the goals of maximizing photovoltaic absorption rate, minimizing grid electricity purchase cost, and ensuring safe battery operation. The SOC safety range, charge / discharge rate limit, and aging stage tolerance of the recombined battery system are used as constraints. Based on the output results of the multi-objective optimization model, the charge / discharge parameters are adjusted in real time. When photovoltaic output is excessive and electricity price is low, the recombined battery system is controlled to charge at the maximum safe rate. During peak grid load and peak electricity price, the battery is discharged to replenish energy. When photovoltaic output fluctuates greatly, the current smoothing coefficient is adjusted to avoid the battery being subjected to frequent power surges. At the same time, the remaining life prediction value is incorporated to balance scenario energy efficiency and long-term battery life, so as to realize the coordinated operation of the recombined battery system and the photovoltaic-storage charging station.

[0078] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of this application.

Claims

1. A method for detecting, sorting, and reassembling waste power batteries, characterized in that, Includes the following steps: S1. Obtain the state parameters of the spent power battery and evaluate the state of charge and health of the spent power battery through an evaluation model that integrates model-driven and data-driven approaches to obtain the state evaluation result. S2. Based on the state assessment results, an adaptive clustering algorithm is used to perform preliminary sorting of the waste power batteries. Then, feature extraction and analysis are used to perform secondary grouping of the pre-sorted waste power batteries to obtain battery packs with consistent matching. S3. Based on the electrical parameter requirements of the photovoltaic-storage charging station scenario, a linear programming algorithm is used to determine the series and parallel configuration of the consistent matching battery packs, and the consistent matching battery packs are recombined to form a recombined battery system. S4. The operating status of the recombinant battery system is monitored in real time by the active equalization battery management system, and the charging and discharging parameters of the recombinant battery system are adjusted in combination with the operating condition fluctuations of the photovoltaic-storage charging station scenario.

2. The method according to claim 1, characterized in that, The evaluation model includes an unscented Kalman filter and a deep neural network. The unscented Kalman filter provides model-driven support, and the deep neural network provides data-driven support.

3. The method according to claim 2, characterized in that, The fusion method of the unscented Kalman filter and the deep neural network is as follows: The deep neural network is trained offline using historical cycle data of spent power batteries to obtain an initial state evaluation model. During online evaluation, the unscented Kalman filter is driven by the collected real-time state parameters to correct the output of the initial state evaluation model in real time, thereby obtaining the state evaluation result.

4. The method according to claim 1, characterized in that, The preliminary sorting of the spent power batteries using an adaptive clustering algorithm specifically includes the following steps: Extract the state parameters of the state of charge and the state parameters of the health state from the state assessment results, and calculate statistical characteristic values; The neighborhood radius of the density clustering algorithm is determined based on the statistical feature values. The neighborhood radius is positively correlated with the dispersion of the state parameters of the charged state and the state parameters of the healthy state. Based on the distribution density of the state parameters of the charged state and the state parameters of the healthy state, the minimum number of samples in the density clustering algorithm is adjusted. The higher the distribution density, the larger the value of the minimum number of samples. The density clustering algorithm is used to perform cluster analysis on waste power batteries, and battery packs with parameter differences that meet the requirements are selected to complete the preliminary sorting.

5. The method according to claim 4, characterized in that, The process of performing secondary grouping of the pre-sorted waste power batteries through feature extraction and analysis specifically includes the following steps: Extract the time series of state parameters of each waste power battery after preliminary sorting; The state parameter time series is transformed into a two-dimensional feature map using the Markov Transition Field, while preserving the correlation characteristics of the state parameters over time. The local window attention mechanism of the Swing Transformer is used to extract the local correlation features of the parameters in the two-dimensional feature map. Combined with the global feature fusion module, the overall distribution law of the parameters is determined to obtain the comprehensive feature vector of each waste power battery. Calculate the similarity between the comprehensive feature vectors of each waste power battery, and group the batteries whose similarity meets the preset conditions into the same group to complete the secondary grouping.

6. The method according to claim 1, characterized in that, After S2 and before S3, it also includes: After secondary grouping and consistency matching, obtain the real-time voltage and capacity data of each waste power battery in the same group, and calculate the voltage deviation and capacity deviation between batteries in the group. Based on the aforementioned state assessment results and the operating condition data of the photovoltaic-storage charging station scenario, the target value for battery balance within the group is determined. Based on the voltage deviation, the capacity deviation, and the equalization target value, the batteries in the group are replenished or released; and during the replenishment or release process, the voltage and capacity changes of each battery in the group are monitored in real time, and the equalization rate is adjusted according to the changing trend until the voltage deviation and capacity deviation of the batteries in the group meet the recombination requirements, thus completing the equalization process.

7. The method according to claim 3, characterized in that, The process of determining the series-parallel configuration of the consistent battery packs using a linear programming algorithm and then reorganizing the consistent battery packs specifically includes the following steps: After equalization processing, obtain the voltage and internal resistance data of each waste power battery, as well as the nominal voltage and nominal capacity of the photovoltaic-storage charging station scenario. An integer linear programming model is constructed, with the number of series and parallel connections in the series-parallel configuration as decision variables, the nominal voltage and nominal capacity as constraints, and minimizing the voltage difference and internal resistance difference within the recombined battery pack as the optimization objective. Solve the integer linear programming model to obtain the optimal series-parallel configuration that satisfies the constraints and achieves the optimization objective; According to the optimal series-parallel configuration, the balanced waste power batteries are connected and assembled to form a recombinant battery system.

8. The method according to claim 7, characterized in that, The condition assessment process also includes predicting the remaining life of the spent power batteries, and the predicted remaining life is used for reconfiguration and adaptation, specifically including: Collect historical charge and discharge cycle data of used power batteries and operating condition data of photovoltaic-storage charging station scenarios; Based on the corrected state parameter values ​​of the state evaluation results from the unscented Kalman filter output, the deep neural network is optimized and trained to construct a remaining life prediction model. The prediction model incorporates the operating condition factors that affect battery life extracted from the operating condition data of the photovoltaic-storage charging station scenario. The remaining life prediction model is used to output the remaining life prediction results for each used power battery. The process of determining the series-parallel configuration and reconfiguration also includes: The remaining lifetime prediction results, along with voltage data and internal resistance data, are used as constraints to ensure that the difference in remaining lifetime among the batteries in the same recombinant battery system meets the preset lifetime requirements.

9. The method according to claim 1, characterized in that, The process of monitoring the operating status of the recombinant battery system in real time through an active balancing battery management system and adjusting the charging and discharging parameters of the recombinant battery system based on the operating data of the photovoltaic-storage charging station scenario specifically includes the following steps: The real-time voltage, real-time internal resistance, and real-time temperature of each individual cell in the recombinant battery system are obtained. Combined with the operating condition data of the photovoltaic-storage charging station scenario, the parameter thresholds and temperature protection thresholds for segmented charging are determined. The charging process employs a segmented charging mode of constant current-constant voltage-trickle charge. During the constant current stage, the current is used to charge the battery to meet the power requirements of the scenario. During the constant voltage stage, the real-time voltage is kept stable within the target voltage range. During the trickle charge stage, the capacity differences between individual battery cells are adjusted and compensated. During the charging process, the active equalization battery management system monitors the real-time temperature of each individual battery cell. When the real-time temperature reaches the temperature protection threshold, the charging rate is reduced. When the real-time temperature exceeds the temperature protection threshold, charging is paused and cooling measures are implemented. Charging is resumed after the real-time temperature drops below the temperature protection threshold.

10. The method according to claim 9, characterized in that, After the reconfigurable battery system completes active balancing management, the following steps are also included: The output voltage and output power response characteristics of the recombinant battery system, as well as the grid connection voltage, frequency standard, and power fluctuations on the photovoltaic side of the power station are obtained. To meet the power grid access standards and reduce grid connection impact, a compatibility adaptation model is constructed. The output voltage of the recombinant battery system, the output power response characteristics, the access voltage of the power grid, the frequency standard, and the power fluctuation on the photovoltaic side are used as inputs to determine the power response adjustment threshold and voltage matching range. Based on the output results of the compatibility adaptation model, the output power response speed of the recombinant battery system is adjusted by the active equalization battery management system, and the power allocation ratio between the photovoltaic side and the battery side is adjusted to reduce the impact of power surges on the power grid.

Images