A method, system, device and storage medium for retired battery hierarchical reorganization based on community discovery and dynamic window strategy

Abstract

The application discloses a kind of based on community discovery and dynamic window strategy's retired battery hierarchical reorganization method, system, equipment and storage medium, belongs to battery echelon utilization technical field.The method is first based on relaxation time distribution Construction battery dynamics fingerprint, utilize jensen-shannon divergence Construction electrochemical similarity network, through community discovery algorithm realizes the precise decoupling of aging mode;Then utilize multilayer perception machine prediction residual capacity, and adopt dynamic greedy sliding window algorithm to carry out module reconstruction, this algorithm is through descending order priority matching and window self-adapting reduction mechanism, while guaranteeing module consistency, significantly improve group rate.The application effectively solves the mechanism mismatch and group rate low double problems brought by the heterogeneity of retired battery, and the system includes test module, graph calculation module, capacity prediction module and grouping decision module.The application also relates to electronic equipment and storage medium for implementing the method.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of retired power battery recycling, cascade utilization and energy storage system integration, and particularly relates to a retired battery grading and reorganization method and system combining graph theory topological analysis and heuristic optimization algorithm, an electronic device and a computer readable storage medium. BACKGROUND

[0002] Under the dual driving of global energy transformation and rapid development of the electric vehicle industry, power lithium-ion batteries are about to usher in a large-scale retirement tide. These batteries still have 70% to 80% of the initial capacity, and if directly recycled for materials, it will cause huge resource waste. Therefore, through cascade utilization, they can be applied to energy storage systems, low-speed electric vehicles, communication base stations and other scenarios, which not only prolongs their life cycle, but also effectively alleviates the pressure of raw material mining, and has great significance for building a sustainable battery recycling economic system. However, the cascade utilization of retired batteries faces serious technical bottlenecks, and the core challenge comes from the significant heterogeneity among individual batteries. This heterogeneity not only exists in the macro indicator of remaining capacity, but also is deeply rooted in the differences in internal aging mechanisms. In the long-term complex use environment, due to different working temperatures, charging and discharging strategies and mechanical stresses, different degrees of active lithium loss, active material loss or interface side reactions may occur inside the battery, leading to differentiation of its electrochemical impedance characteristics. If only according to the capacity, the traditional "capacity grading" grouping is easy to combine batteries with similar capacity but completely different aging paths. For example, a battery with capacity attenuation due to negative electrode lithium precipitation and a battery with capacity attenuation due to positive electrode material crushing, although their current capacities are the same, their performance evolution, voltage response and thermal safety in subsequent cycles will be very different. When these mismatched batteries are connected in series or parallel, serious current and voltage distribution imbalance will be caused in the charging and discharging process. The energy of high-capacity batteries cannot be fully utilized, while the "short board" batteries will reach the voltage cutoff condition or produce internal circulation in advance, which not only leads to a sharp decrease in the available capacity of the whole group, but also causes additional energy loss and local overheating, sharply accelerating the aging process of all batteries and burying potential safety hazards such as thermal runaway.

[0003] Currently, the gold standard for obtaining battery capacity is the full charge-discharge test. While this method is highly accurate, it takes several hours, which cannot meet the efficiency requirements for rapid screening of large-scale retired batteries. To accelerate this process, the industry has begun to explore using easily measurable external features, such as open-circuit voltage and ohmic internal resistance, to indirectly predict capacity. However, the relationship between open-circuit voltage, ohmic internal resistance, and remaining capacity is subject to complex coupling effects from various factors such as battery chemistry, aging path, and temperature, exhibiting strong nonlinearity. This limits the accuracy of simple linear regression models. Furthermore, even after obtaining accurate capacity, efficiently pairing hundreds or thousands of batteries into highly consistent modules is itself a complex combinatorial optimization problem. The fixed-number sliding window algorithm commonly used in existing technologies is too rigid when dealing with the large dispersion of retired battery capacity distribution. Once the capacity difference within a candidate window exceeds a preset threshold, the entire window of batteries is often skipped or downgraded. This "either / or" strategy results in a large number of batteries with acceptable performance being wasted because they cannot be matched into a fixed number, severely reducing the overall assembly rate and the economic benefits of tiered utilization.

[0004] In summary, there is an urgent need in this field for an innovative technical solution that can penetrate macroscopic appearances, achieve accurate clustering at the microscopic mechanism level, and possess high flexibility at the macroscopic grouping level, thereby systematically solving the problem of heterogeneity in retired batteries. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method, system, device, and storage medium for graded reorganization of retired batteries based on community discovery and dynamic window strategies. This addresses the problems of mismatched module dynamic performance caused by ignoring differences in battery aging mechanisms and low grouping rates due to fixed grouping strategies. To achieve the above objectives, this invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for graded reorganization of retired batteries based on community discovery and dynamic window strategy, comprising the following steps:

[0007] S1. Obtain the relaxation time distribution of each battery cell in the retired battery cluster. Spectrum, calculate any two battery cells Jensen-Shannon divergence between spectra Based on the above Construct a sparse weighted network graph for battery cells;

[0008] S2. Application The community detection algorithm performs community partitioning on the sparse weighted network graph, dividing the battery cells into multiple battery clusters with similar internal electrochemical characteristics;

[0009] S3. Measure open circuit voltage of each battery cell and ohmic internal resistance , the and input pre-trained multi-layer perceptron model to get the predicted remaining capacity of each battery cell, and sort the battery cells in each of the battery clusters in descending order of the predicted remaining capacity;

[0010] S4. For the sorted battery cell list, use a dynamic greedy sliding window algorithm for module reconstruction, where the target module battery number , the minimum allowed battery number and the maximum allowed capacity range threshold are set, and the algorithm is iteratively executed from the start of the list, and each iteration process includes:

[0011] S41. Initialize the window size to , covering consecutive battery cells starting from the current pointer position;

[0012] S42. Calculate the difference between the maximum predicted capacity and the minimum predicted capacity of the battery cells in the current window ;

[0013] S43. If , divide the battery cells in the current window into a battery module, and move the list processing pointer forward by positions;

[0014] S44. If , reduce the window size by 1, repeat steps S42 and S43 until a battery cell combination that meets or the window size is less than is found.

[0015] As an optional implementation, the step S1 of constructing a sparse weighted network graph includes:

[0016] normalizing the spectrum graph vector of each battery cell to obtain the corresponding probability distribution ;

[0017] For any two battery cells and , calculate the Jensen-Shannon divergence between them according to the following formula :

[0018] ;

[0019] wherein, and respectively represent the battery unit and the battery unit ; the probability distribution of the normalized spectrum vector, represents and the mean distribution of, represents the relative entropy;

[0020] define the edge weight between the battery unit and the battery unit ;

[0021] Adopt -neighbor strategy, for each battery unit node, only keep the highest edge weight of the edges, forming the sparse weighted network graph.

[0022] As an optional implementation, in the step S2, the community discovery algorithm optimizes the modularity by iteratively performing the following stages:

[0023] Local movement stage: traverse the nodes in the network graph, and move each node to the neighbor community that can maximize the modularity gain of the community to which it belongs;

[0024] Community refinement stage: after local movement, each node is regarded as an independent sub-community within the initial community formed, and only allows nodes to merge with neighbor nodes that belong to the same initial community and are connected, forming connected sub-communities;

[0025] Network aggregation stage: aggregate the refined sub-communities into new super nodes, and repeat the above process based on the aggregated network until the modularity no longer significantly improves.

[0026] As an optional implementation, the calculation formula of the modularity is as follows:

[0027] ;

[0028] wherein, is the sum of the weights of all edges in the network graph, is the edge weight between the node and the node , and are respectively the weighted degrees of the node and the node , and They are nodes and nodes The community to which it belongs Let Kronecker function be used when = It is 1 if it is true, otherwise it is 0.

[0029] As an optional implementation, the pre-training in step S3 The model includes:

[0030] The input layer has two neurons that receive the normalized input signal. and data;

[0031] At least one hidden layer, each hidden layer containing multiple neurons and employing a non-linear activation function;

[0032] The output layer has one neuron that outputs the predicted remaining capacity value;

[0033] Among them, the The model is trained through supervised learning, and the training data includes sample batteries. , And its actual capacity value obtained through standard full-fill and full-discharge tests.

[0034] In a second aspect, the present invention provides a retired battery grading and regeneration system for implementing the method described in the first aspect, comprising:

[0035] The test module is used to measure the open-circuit voltage of the battery cell. and Ohmic internal resistance ;

[0036] The graph calculation module, which is communicatively connected to the test module, is used to execute... Transform, construct, and run electrochemical similarity networks The community detection algorithm outputs battery cluster tags;

[0037] The capacity prediction module is communicatively connected to the test module and has the pre-trained [database / system] built-in. Model, used to determine and Predict the remaining capacity of the battery cells;

[0038] The group decision module is communicatively connected to the graph calculation module and the capacity prediction module, and incorporates the dynamic greedy sliding window algorithm to output battery module configuration schemes.

[0039] As an optional implementation, the test module includes:

[0040] High precision voltmeter for measuring open circuit voltage of battery cells ;

[0041] Internal resistance tester for measuring ohmic internal resistance of battery cells .

[0042] In a third aspect, the present application provides an electronic device comprising:

[0043] one or more processors;

[0044] a memory;

[0045] one or more computer programs;

[0046] wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and the one or more computer programs contain instructions for performing the method of the first aspect.

[0047] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method of the first aspect.

[0048] The present application introduces an electrochemical similarity network based on relaxation time distribution and The community discovery algorithm first realizes the accurate decoupling of the battery aging mode from the micro mechanism level, thereby ensuring that the batteries in the same cluster not only have similar capacities, but also have highly consistent intrinsic electrochemical impedance characteristics and degradation paths, which fundamentally eliminates the differentiation of module dynamic performance and the "short board effect" caused by mechanism mismatch. On this basis, the dynamic greedy sliding window algorithm breaks the rigid mode of traditional fixed number grouping through its unique window adaptive reduction mechanism, realizes the optimization trade-off of "exchanging consistency with scale", and recombines those batteries that have been discarded due to slightly poor performance into qualified modules with slightly smaller scale but consistent standards, thereby significantly improving the resource utilization and grouping rate of retired batteries. Moreover, the entire scheme deeply integrates electrochemical atlas analysis, graph theory community discovery, machine learning prediction and combinatorial optimization algorithm. This cross-disciplinary technical cooperation realizes the whole-process optimization of retired batteries from mechanism identification to final grouping without significantly increasing the computational complexity.

[0049] In summary, the present application provides an efficient, automatic and configurable systematic solution, which not only greatly improves the consistency guarantee of recombination modules in terms of cycle life and safety, but also directly enhances the economic feasibility of the cascade utilization project through a high grouping rate, providing a solid technical support for the high-value and large-scale utilization of retired batteries. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced.

[0051] Figure 1 The overall flowchart of the grouping method provided in the embodiments of the present application is shown in the figure.

[0052] Figure 2 The overall framework and data flow direction of the grouping method in the embodiments of the present application are shown in the figure.

[0053] Figure 3 The performance of the community discovery algorithm in the embodiments of the present application is shown in the figure, wherein:

[0054] Figure 3 (a) in the figure shows the relaxation time distribution of four clusters (T1, T2, T3 and T4)

[0055] Figure 3 (b) in the figure shows the voltage curve and the differential capacity (C / 2) curve of the charging process of the battery in the four clusters;

[0056] Figure 3 (c) in the figure shows the voltage curve and the differential capacity (C / 2) curve of the discharging process of the battery in the four clusters;

[0057] Figure 4 The correlation analysis diagram between the key electrochemical parameters is shown in the figure.

[0058] Figure 5 The performance diagram of the greedy grouping algorithm in the embodiments of the present application is shown in the figure.

[0059] Figure 5 (a) in the figure shows the comparison scatter diagram of the predicted capacity and the real capacity;

[0060] Figure 5 (b) in the figure shows the mapping relationship cloud diagram / contour diagram of the open circuit voltage, ohmic internal resistance and residual capacity;

[0061] Figure 5 (c) in the figure shows the real capacity, coulomb efficiency (ηC) , voltage efficiency (ηV) and energy efficiency (ηE) of the member batteries in each module finally formed in the form of a bee swarm diagram;

[0062] Figure 6 The cluster group 1 (T1) ​​​​​The battery module partitioning results in the image show the relaxation time distribution of each battery. Curves and dynamic cycle curves;

[0063] Figure 7 For cluster 1 ( The relaxation process curve of each cell in the battery module;

[0064] Figure 8 For cluster 2 ( The battery module partitioning results in the diagram show the configuration of each battery. Curves and dynamic cycle curves;

[0065] Figure 9 For cluster 2 ( The relaxation process curve of each cell in the battery module;

[0066] Figure 10 For cluster 3 ( The battery module partitioning results in the diagram show the configuration of each battery. Curves and dynamic cycle curves;

[0067] Figure 11 For cluster 3 ( The relaxation process curve of each cell in the battery module;

[0068] Figure 12 For cluster 4 ( The battery module partitioning results in the diagram show the configuration of each battery. Curves and dynamic cycle curves;

[0069] Figure 13 For cluster 4 ( The relaxation process curve of each battery in the battery module. Detailed Implementation

[0070] Example 1:

[0071] This embodiment provides a method for graded reorganization of decommissioned batteries based on community discovery and dynamic window strategy. The overall process is as follows: Figure 1 As shown, the methodology comprises four core stages: relaxation time distribution analysis, community detection model, multilayer perceptron model, and greedy optimization using dynamic windows. The overall framework and data flow can be found in [reference needed]. Figure 2 As shown, it fully demonstrates the process from raw battery data to graph structure construction, then to mechanistic clustering and final group output.

[0072] First, the first key step of the method is S1, constructing the electrochemical similarity network. To ensure the consistency of the battery from the micro-mechanism level, this embodiment does not use simple capacity or internal resistance as the only basis for grouping. Instead, this embodiment uses electrochemical impedance spectroscopy tests to obtain the relaxation time distribution of each battery cell in the retired battery cluster to be reorganized by solving the first kind of Fredholm integral equation . The spectrum is regarded as the "dynamic fingerprint" of the battery, which can effectively characterize the electrochemical processes inside the battery, such as active lithium loss, active material loss or ohmic impedance increase, thereby achieving deep decoupling of the battery aging mode.

[0073] Next, in order to quantify the electrochemical similarity between any two battery cells, it is necessary to calculate the Jensen-Shannon divergence between the spectrum of any two battery cells.

[0074] First, normalize the spectrum vector of each battery cell to obtain the corresponding probability distribution .

[0075] Second, according to the calculation formula of , calculate the Jensen-Shannon divergence between the battery cell and . The formula is:

[0076] ;

[0077] wherein and respectively represent the probability distribution of the spectrum vector of the battery cell after normalization, represents the mean distribution of and , and represents the relative entropy. The value reflects the difference between the two spectrum probability distributions, and the smaller the difference, the higher the similarity. Third, define the edge weight

[0078] between the battery cell and , which is converted based on , and the specific formula is:

[0079] ​.

[0080] Obviously, the greater the value of similarity weight , the more similar the electrochemical characteristics of the two batteries.

[0081] Fourthly, the algorithm adopts - a near-neighbor strategy, for each battery node, only the edges with the highest edge weight are kept, and the rest of the edge weights are optionally set to zero. For example, the value can be selected as 10, thus forming the sparse weighted network graph. This sparse graph structure avoids the huge computational complexity brought by the fully connected graph, while retaining the most important electrochemical similarity connections.

[0082] Then, the second key step of the method is S2, community detection algorithm divides the sparse weighted network graph into communities. The algorithm is an advanced graph topology analysis tool, aiming to divide the battery units into multiple battery clusters with similar internal electrochemical characteristics. The algorithm optimizes the modularity of the graph by iteratively performing three stages: local movement stage, community refinement stage, and network aggregation stage. .

[0083] In the local movement stage, the algorithm traverses each node in the network graph and moves each node to the neighbor community that maximizes the modularity gain of the community it belongs to. This stage is a preliminary community aggregation process.

[0084] Then comes the community refinement stage, which is also the key innovation point of the algorithm compared to traditional algorithms. After the local movement, the algorithm does not directly merge all nodes in the initial community into one whole. Instead, it treats each node as an independent sub-community and only allows the node to merge with neighbor nodes that also belong to the initial community and are closely connected. By limiting the scope of merging, it ensures that the generated sub-communities are topologically connected. This mechanism mathematically excludes those abnormal battery nodes that are "classified as a category, but actually hang on the edge of the community through weak connections", thus avoiding the intra-community disconnection problem that may occur in algorithms, ensuring that the batteries within the cluster have high consistency and homogeneity in electrochemical mechanisms.

[0085] Finally, the network aggregation stage, the algorithm aggregates the refined sub-communities into new super-nodes, and repeats the above process based on the aggregated network until the modularity no longer significantly improves, reaching a state of convergence.

[0086] In this process, the modularity It is a core indicator for measuring the quality of community zoning, and its calculation formula is as follows:

[0087] ;

[0088] in, This is the sum of the weights of all edges in the network graph. For nodes and nodes Edge weights between them and They are nodes and nodes The weighting degree. and They are nodes and nodes The community to which it belongs Let Kronecker function be used when = It is 1 if it is true, and 0 otherwise. This module degree can be optimized. Ultimately, all batteries were divided into several battery clusters with similar aging mechanisms.

[0089] For example, in a company containing 187 commercialized When validating on a dataset of cylindrical batteries, The algorithm automatically divided these 187 batteries into 4 clusters. For example... Figure 3 As shown, the impedance spectrum shapes of cells within the same cluster highly overlap. Figure 3 (a) clearly shows the four clusters ( to relaxation time distribution Spectrum. Figure 3 (b) and Figure 3 (c) shows the charging voltage curves and differential capacity of the batteries in these four clusters, respectively. Curves, and the voltage curve and differential capacity during the discharge process. The high degree of consistency among these curves visually demonstrates the passage of time. The effectiveness of the algorithm in performing mechanistic clustering.

[0090] Furthermore, the third key step of the method is S3, capacity prediction and ranking. After completing the mechanistic clustering, for each cell within each cluster, it is necessary to measure its macroscopic health status indicators. Specifically, this step requires measuring the open-circuit voltage of each cell. and Ohmic resistance To ensure measurement accuracy, the measurement can optionally be performed after adjusting the battery to a uniform level, for example, 20%. The process is carried out after the state is stabilized and left to stand still.

[0091] Subsequently, the aforementioned and As input features, they are input into the pre-trained multilayer perceptron. The model is designed to quickly obtain the predicted remaining capacity value for each battery cell. This pre-trained MLP model includes:

[0092] The input layer, which optionally has two neurons, receives the normalized input... and data;

[0093] At least one hidden layer, each containing multiple neurons and employing a non-linear activation function, may optionally include a batch normalization layer and... layer;

[0094] And an output layer, which has a single neuron that outputs the predicted remaining capacity value.

[0095] For example, a method with two hidden layers can be used. The model has a first hidden layer that optionally contains 16 neurons and a second hidden layer that optionally contains 8 neurons. Each hidden layer is optionally followed by a... The activation function, a batch normalization layer, and a dropout rate optionally of 0.2 are used. layer.

[0096] Should The model is trained using supervised learning, and the training data includes sample batteries. , It also includes the actual capacity value obtained through a standard full-fill-discharge test. The training objective is to minimize the mean square error between the predicted capacity and the actual capacity. Figure 4 The correlation analysis of key electrochemical parameters shows that... and The main characterization stage is related to the health status. It has a high negative correlation ( =-0.92 and =-0.84), which provides an intrinsic relational basis for using them as input features for capacity regression prediction.

[0097] Should The model exhibits extremely high predictive performance on the validation set. For example... Figure 5 As shown, Figure 5 The scatter plot comparing the predicted and actual values ​​in (a) clearly shows that the data points are closely distributed on the ideal plane. Mean absolute percentage error around the diagonal The root mean square percentage error is 0.88%. The coefficient of determination is 1.17%. Up to 0.9221. Figure 5 (b) in the diagram visually illustrates the open-circuit voltage. Ohmic internal resistance The mapping relationship between the cloud map and the remaining capacity, or contour map, illustrates the mapping capability of the prediction model. In practical applications, for any battery cluster to be grouped, it is only necessary to quickly measure the capacity of each battery within the cluster. and This pre-trained [system] can be used The model can predict the remaining capacity of all batteries within seconds.

[0098] After obtaining the predicted capacity, the final step in this process is to sort the battery cells within each battery cluster in descending order of their predicted remaining capacity. The purpose of descending order sorting is to implement a greedy strategy of "descending order priority matching," prioritizing and retaining high-capacity, high-value batteries.

[0099] Finally, the fourth key step of the method is S4 dynamic window fine-tuning grouping. For the sorted list of battery cells, a dynamic greedy sliding window algorithm is used for module reconstruction. This step aims to address the low grouping rate and resource waste caused by traditional fixed grouping strategies.

[0100] This algorithm requires setting three core parameters: the number of batteries in the target module. Minimum allowed number of batteries and the maximum allowable capacity range threshold For example, in an application where the target assembly is a 14-cell module (48V), it can be set... It can be 14, with a maximum permissible capacity range. 100mAh, minimum allowable window size For 5 strings or less.

[0101] The algorithm iterates from the beginning of the sorted list. Each iteration includes the following sub-steps:

[0102] S41, Initialize window size for Covers the continuous range starting from the current pointer position. One battery cell.

[0103] S42, calculate the current window's... The difference between the maximum and minimum predicted capacity of each battery cell . Capacity variation is a hard constraint indicator for consistency within a module.

[0104] S43, perform the range constraint test. If the calculated satisfies , the consistency condition is satisfied. The current window of battery units is divided into a battery module, and the list processing pointer is moved forward by bits to start the next iteration. For example, if = 14, = 100 mAh, and the calculated range is 45 mAh, the grouping is successful, and the pointer is moved forward by 14 bits.

[0105] S44, if , it means that the current window content is not consistent enough. At this time, the algorithm does not give up, but performs window reduction logic. The window size is reduced by 1, and optionally, the battery with the smallest predicted capacity in the current window is removed because it is usually the "short board" that causes the range to be too large. Then, the capacity range of the remaining -1 batteries is recalculated , steps S42 and S43 are repeated until a battery unit combination that satisfies is found, or the window size is smaller than the preset minimum allowed value . If a sub-group that satisfies the condition is found, the sub-group is output as an independent module, and the pointer is moved back by bits. If the window size is smaller than and still cannot be grouped, this iteration ends, and the pointer is optionally moved back by only 1 bit to try to match again starting from the battery with the second highest capacity.

[0106] This dynamic greedy sliding window algorithm realizes the flexible grouping strategy of "sacrificing module size for consistency" through the "window self-adaptive reduction" mechanism. For example, in the second iteration, if the initial window = 14 has a range of , and the calculation result is 112 mAh, which exceeds the threshold of 100 mAh. The algorithm reduces the window size to 13, recalculates the range, and if the assumption is 88 mAh, the condition is satisfied. The 13 batteries are successfully divided into the second module, and the pointer is moved forward by 13 bits. This process is repeated until all batteries are assigned to the corresponding module.

[0107] As shown in the swarm plot in (c) of Figure 5 , it clearly shows the real capacity, coulomb efficiency , voltage efficiency and energy efficiency of the member batteries inside each module (horizontal coordinate is module number, e.g. 1-1, 1-2, etc.)All exhibited extremely high clustering and consistency. This intuitively demonstrates the effectiveness of the refined grouping method proposed in this embodiment in terms of macroscopic capacity consistency.

[0108] Figures 6 to 13 Further detailed electrochemical characteristics of the grouped modules in different clusters are presented. These results collectively demonstrate that, by first based on Perform mechanistic clustering, and then through - Dynamic windows are used to finely group the capacity, and the resulting modules are not only highly consistent in macroscopic capacity, but also have extremely high homogeneity in microscopic electrochemical characteristics.

[0109] In summary, the method provided in Embodiment 1, by using a method based on Electrochemical fingerprints and By combining community detection algorithms, precise decoupling of battery aging modes is achieved at the microscopic mechanism level, ensuring high homogeneity of batteries within the cluster. Subsequently, it is combined with... The millisecond-level capacity prediction and dynamic greedy sliding window algorithm significantly improve the packing rate at the macroscopic capacity level. This two-stage hierarchical reorganization architecture effectively solves the packing problem caused by the heterogeneity of retired batteries, significantly improving the economic benefits and overall utilization rate of tiered utilization while ensuring the consistency of module performance. The entire method does not require complex physical modeling or long-term charge-discharge testing, and has high efficiency and robustness, making it suitable for the industrial sorting needs of large-scale retired batteries.

[0110] Example 2:

[0111] This embodiment provides a retired battery grading and reorganization system for implementing the method described in Embodiment 1 above. The retired battery grading and reorganization system includes: a testing module, a graph calculation module, a capacity prediction module, and a grouping decision module. Each module is interconnected within the system and works collaboratively to implement the grading and reorganization method.

[0112] Test Module: This module is the system's physical interface, used to acquire key characteristic parameters of the battery. It is used to measure the open-circuit voltage of the battery cells. and Ohmic internal resistance The testing module optionally includes a high-precision voltmeter and an internal resistance tester. The high-precision voltmeter is used to measure the open-circuit voltage of the battery cell in a static state. The internal resistance tester is used to measure the ohmic internal resistance of the battery cell. For example, it can be achieved through the AC impedance method or Test acquisition. Furthermore, in order to obtain the necessary data for constructing electrochemical similarity networks... Spectra, this module also needs to have Test functions and Transformation and computation capabilities.

[0113] Graph computing module: This module is communicatively connected with the test module, and is configured to perform transformations, based on constructing electrochemical similarity networks, and running community discovery algorithms. The computing core of the graph computing module is to iteratively perform three stages of local movement, community refinement and network aggregation to maximize modularity , and finally output battery cluster labels with similar aging mechanisms. The computing result of this module is the mechanism clustering of the battery aging mode, which lays a foundation for subsequent capacity fine grouping.

[0114] Capacity prediction module: This module is also communicatively connected with the test module, and has a pre-trained multi-layer perception model built-in. The model is configured to quickly predict the remaining capacity of the battery unit according to the and data provided by the test module. The architecture and training process of this module strictly follow the description in Embodiment One, ensuring millisecond-level prediction speed and extremely high accuracy. The output of the capacity prediction module is the predicted remaining capacity value of each battery unit.

[0115] Grouping decision module: This module is communicatively connected with the graph computing module and the capacity prediction module, and is the core component for realizing dynamic optimization grouping. It has the dynamic greedy sliding window algorithm built-in, which is configured to receive the battery cluster labels from the graph computing module and the predicted capacity values from the capacity prediction module. This module first sorts the batteries in each battery cluster in descending order of predicted capacity, and then performs step S4, including window initialization, capacity range checking, and grouping and pointer movement when the condition is met, and window adaptive reduction logic when the condition is met. This module finally outputs the battery module configuration scheme after hierarchical reorganization, for example, which battery unit IDs are included in each module. The grouping decision module can also be communicatively connected with the input / output module to receive the battery information to be tested and output the final battery module grouping result.

[0116] During system operation, the test module is responsible for collecting physical data, and transmitting data and and data to the graph computing module and the capacity prediction module. The graph computing module processes and clusters the data, determines the battery cluster attribution, and sends the battery cluster labels to the grouping decision module. The capacity prediction module processes and predicts the and data, and outputs the predicted remaining capacity value of each battery unit.The data is used for capacity prediction, and the predicted capacity value is sent to the grouping decision module. The grouping decision module comprehensively utilizes the mechanism cluster tags and the predicted capacity, executes a dynamic greedy grouping algorithm, and generates and outputs the final module configuration scheme. The entire system, through modular design, effectively integrates mechanism analysis, rapid prediction, and flexible grouping, ensuring the efficiency and quality of graded reorganization of retired batteries.

[0117] Example 3:

[0118] This embodiment provides an electronic device, including one or more processors, a memory, and one or more computer programs. The one or more computer programs are stored in the memory and configured to be executed by the one or more processors. The one or more programs contain instructions for performing the method described in Embodiment 1, that is, when the processor executes these instructions, it is able to perform... Spectral analysis, calculate, Community discovery, All methods and steps, including capacity prediction, descending sorting, and dynamic greedy sliding window algorithm.

[0119] The electronic device may optionally be a control computer, embedded controller, or high-performance server for an industrial battery sorting line.

[0120] Example 4:

[0121] This embodiment provides a computer-readable storage medium storing a computer program or computer instructions thereon. When executed by a processor, the computer program or instructions can implement the steps of the method described in Embodiment 1. The computer-readable storage medium may optionally include a non-transient storage medium, such as a read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive, solid-state drive, etc. The instructions can be configured to implement all the computational and logical judgment processes in the hierarchical reconfiguration method, thereby enabling the processor to process retired battery data into a final module configuration scheme.

[0122] In summary, the retired battery tiered recombination system disclosed in Embodiment 2 fully implements the two-stage tiered recombination method described in Embodiment 1 by functionally dividing and communicating the testing module, graph calculation module, capacity prediction module, and grouping decision module. This system architecture not only supports clustering based on micro-mechanisms but also achieves rapid and refined grouping based on macro-capacity. Presented in the form of the electronic devices and computer-readable storage media described in Embodiments 3 and 4, this system provides the battery tiered utilization industry with an efficient, robust, and industrially implementable integrated hardware and software solution, significantly improving the comprehensive utilization rate and economic benefits of retired batteries.

Claims

1. A method for graded reorganization of decommissioned batteries based on community detection and dynamic window strategy, characterized in that, Includes the following steps: S1. Obtain the relaxation time distribution of each battery cell in the retired battery cluster. Spectrum, calculate any two battery cells Jensen-Shannon divergence between spectra Based on the above Construct a sparse weighted network graph for battery cells; S2. Application The community detection algorithm performs community partitioning on the sparse weighted network graph, dividing the battery cells into multiple battery clusters with similar internal electrochemical characteristics; S3. Measure the open-circuit voltage of each battery cell. and Ohmic resistance , will the and Input pre-trained multilayer perceptron The model obtains the predicted remaining capacity of each battery cell, and within each battery cluster, the battery cells are sorted in descending order according to the predicted remaining capacity. S4. For the sorted list of battery cells, a dynamic greedy sliding window algorithm is used for module reconstruction, wherein the target module battery quantity is set. Minimum number of batteries allowed and the maximum allowable capacity range threshold The algorithm iterates from the beginning of the list, and each iteration includes: S41. Initialize window size for Covers the continuous range starting from the current pointer position. One battery cell; S42. Calculate the difference between the maximum predicted capacity and the minimum predicted capacity of the battery cells within the current window. ; S43. If Then the current window will contain Each battery cell is divided into a battery module, and the list processing pointer is moved forward. Bit; S44. If If the window size is not found, reduce the window size by 1 and repeat steps S42 and S43 until a satisfactory result is found. Battery cell combination or window size Less than .

2. The method for graded reorganization of decommissioned batteries based on community discovery and dynamic window strategy according to claim 1, characterized in that, The construction of the sparse weighted network graph in step S1 includes: For each battery cell The spectral vectors are normalized to obtain the corresponding probability distribution. ; For any two battery cells and The Jensen-Shannon divergence between the two can be calculated using the following formula. : ； in, and Representing battery cells and battery cells of The normalized probability distribution of the spectral vector. express and The mean distribution, Represents relative entropy; Define battery cell and Edge weights between ; use - The nearest neighbor strategy retains only the node with the highest edge weight for each battery cell node. Edges are formed to create the sparse weighted network graph.

3. The method for graded reorganization of decommissioned batteries based on community discovery and dynamic window strategy according to claim 1, characterized in that, In step S2 The community detection algorithm optimizes modularity by iteratively executing the following stages. : Local move phase: Traverse the nodes in the network graph and move each node to a neighboring community that maximizes the modularity gain of its own community; Community refinement stage: Within the initial community formed after local migration, each node is regarded as an independent sub-community, and only nodes are allowed to merge with neighboring nodes that belong to the same initial community and are connected to form connected sub-communities; Network aggregation phase: The refined sub-communities are aggregated into new supernodes, and the above process is repeated based on the aggregated network until the modularity is reached. It no longer improves significantly.

4. The method for graded reorganization of decommissioned batteries based on community discovery and dynamic window strategy according to claim 3, characterized in that, modularity The calculation formula is: ； in, This is the sum of the weights of all edges in the network graph. For nodes and nodes Edge weights between them and They are nodes and nodes The weighting degree; and They are nodes and nodes The community to which it belongs Let Kronecker function be used when = It is 1 if it is true, otherwise it is 0.

5. The method for graded reorganization of decommissioned batteries based on community discovery and dynamic window strategy according to claim 1, characterized in that, The pre-training in step S3 The model includes: The input layer has two neurons that receive the normalized input signal. and data; At least one hidden layer, each hidden layer containing multiple neurons and employing a non-linear activation function; The output layer has one neuron that outputs the predicted remaining capacity value; Among them, the The model is trained through supervised learning, and the training data includes sample batteries. , And its actual capacity value obtained through standard full-fill and full-discharge tests.

6. A retired battery grading and reconfiguration system for implementing the method according to any one of claims 1 to 5, characterized in that, include: The test module is used to measure the open-circuit voltage of the battery cell. and Ohmic resistance ; The graph calculation module, which is communicatively connected to the test module, is used to execute... Transform, construct, and run electrochemical similarity networks The community detection algorithm outputs battery cluster tags; The capacity prediction module is communicatively connected to the test module and has the pre-trained [database / system] built-in. Model, used to determine and Predict the remaining capacity of the battery cells; The group decision module is communicatively connected to the graph calculation module and the capacity prediction module, and incorporates the dynamic greedy sliding window algorithm to output battery module configuration schemes.

7. The decommissioned battery grading and recombination system according to claim 6, characterized in that, The testing module includes: High-precision voltmeter for measuring the open-circuit voltage of battery cells. ; An internal resistance tester is used to measure the ohmic internal resistance of a battery cell. .

8. An electronic device, characterized in that, include: One or more processors; Memory; One or more computer programs; The one or more computer programs are stored in the memory and configured to be executed by the one or more processors, the one or more computer programs containing instructions for performing the method as described in any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 5.

Images