Data-driven method, device, equipment and medium for sorting retired batteries

Abstract

This application discloses a data-driven method, apparatus, equipment, and medium for sorting retired batteries, relating to the field of retired battery sorting. It extracts discharge voltage curves characterizing the battery's voltage plateau, polarization resistance, and other properties during operation from lithium-ion battery variation data, avoiding the limitations of single-point parameters. An optimized t-SNE algorithm is used to reduce the dimensionality of the discharge voltage curves. This optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm, solving the problem of poor dimensionality reduction caused by blindly setting parameters in traditional t-SNE, making low-dimensional data both highly efficient in compression and retaining the discriminative power of key features. An improved K-means clustering strategy based on a local gravity model is used to classify the dimensionality-reduced voltage curves into multiple performance category groups, eliminating noise interference at the source and preventing cluster center shift. The multiple performance category groups are evaluated, significantly improving the efficiency and accuracy of retired battery sorting.

Description

Technical Field

[0001] This application relates to the field of retired battery sorting technology, and in particular to a data-driven retired battery sorting method, apparatus, equipment and medium. Background Technology

[0002] Lithium-ion batteries are core energy storage components in electric vehicles, energy storage systems, and other fields. Their efficient recycling and reuse after retirement is crucial for achieving resource recycling and promoting green development. Rapid and accurate sorting of retired batteries is a fundamental prerequisite for ensuring their performance and safety during secondary use. As service time increases, lithium-ion batteries exhibit significant performance differences due to factors such as charge-discharge cycles and storage environments. Parameters such as capacity decay, polarization resistance, and voltage plateau characteristics vary. Direct mixing and recombination can easily lead to poor battery pack consistency, shortened cycle life, and even safety hazards such as thermal runaway. Therefore, achieving accurate classification of retired batteries through efficient sorting methods has significant engineering value for improving recycling efficiency and reducing the risks of secondary applications.

[0003] In related technologies, current methods for sorting retired batteries mainly fall into two categories: one is the traditional detection method based on physical parameters, which sorts batteries by directly measuring single-point parameters such as capacity and internal resistance; the other is the data-driven intelligent sorting method, which uses the battery's complete discharge voltage curve as a sorting feature to reflect the battery's characteristics such as voltage plateau, ohmic internal resistance, and polarization internal resistance during operation. However, the applicant recognizes that the first method relies only on local features, making it difficult to comprehensively reflect the overall performance differences of the batteries, resulting in limited sorting accuracy and a long detection process, which is unsuitable for processing large-scale retired batteries. The performance of the second method is highly dependent on the sufficiency of feature extraction, optimization of parameters such as the perplexity of t-SNE and the learning rate, as well as the clustering effect of the clustering algorithm. However, problems such as unreasonable dimensionality reduction parameter settings and susceptibility to noise interference in traditional methods make it difficult to meet the actual needs in terms of sorting efficiency and accuracy, thus restricting the large-scale and efficient recycling of retired batteries. Summary of the Invention

[0004] In view of this, this application provides a data-driven method, apparatus, equipment and medium for sorting retired batteries. The main purpose is to solve the problems of the first method, which has limited sorting accuracy and efficiency due to reliance on local features; and the second method, which is affected by t-SNE parameters, clustering algorithms, etc., and traditional schemes have problems such as unreasonable dimensionality reduction parameters and clustering being susceptible to noise interference, resulting in insufficient sorting efficiency and accuracy, which restricts the large-scale and efficient recycling of retired batteries.

[0005] According to a first aspect of this application, a data-driven method for sorting decommissioned batteries is provided, the method comprising: Acquire lithium-ion battery variation data and extract the discharge voltage curve from the lithium-ion battery variation data; The discharge voltage curve is reduced in dimension by the optimized t-SNE algorithm to obtain the dimension-reduced voltage curve. The optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm. The reduced voltage curves are classified using an improved K-means clustering algorithm to obtain multiple performance category groupings. The improved K-means clustering algorithm is obtained by improving the K-means clustering algorithm using the grid partitioning strategy of the local gravity model. The data grouped by the multiple performance categories are evaluated to obtain the sorting results of retired batteries.

[0006] According to a second aspect of this application, a data-driven decommissioned battery sorting device is provided, the device comprising: The acquisition module is used to acquire lithium-ion battery change data and extract the discharge voltage curve from the lithium-ion battery change data. The dimension reduction module is used to perform dimension reduction processing on the discharge voltage curve using the optimized t-SNE algorithm to obtain the dimension-reduced voltage curve. The optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm. The classification module is used to classify the dimension-reduced voltage curves using an improved K-means clustering algorithm to obtain multiple performance category groupings. The improved K-means clustering algorithm is obtained by improving the K-means clustering algorithm using the grid partitioning strategy of the local gravity model. The evaluation module is used to evaluate the grouped data of the multiple performance categories to obtain the sorting results of retired batteries.

[0007] According to a third aspect of this application, an apparatus is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described in any of the first aspects above.

[0008] According to a fourth aspect of this application, a medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the first aspects above.

[0009] By employing the above technical solutions, the technical solutions provided in the embodiments of this application have at least the following advantages: This application provides a data-driven method, apparatus, equipment, and medium for sorting retired batteries. This application extracts discharge voltage curves characterizing the battery's voltage plateau, polarization resistance, and other properties during operation from lithium-ion battery variation data. This reflects battery performance across the entire voltage cycle, avoiding the limitations of single-point parameters and providing a reliable basis for sorting. Next, the discharge voltage curves are dimensionality-reduced using an optimized t-SNE algorithm. The optimized t-SNE algorithm is obtained by optimizing the traditional t-SNE algorithm using the IWMA algorithm, solving the problem of poor dimensionality reduction caused by blindly setting parameters in the traditional t-SNE algorithm. This ensures efficient compression of low-dimensional data while retaining the discriminative power of key features. Subsequently, the dimensionality-reduced voltage curves are classified using an improved K-means clustering algorithm, resulting in multiple performance category groups. The improved K-means clustering algorithm is derived by using a grid partitioning strategy based on a local gravity model, eliminating noise interference at the source, preventing cluster center shift, and allowing retired batteries with different performance characteristics to be accurately grouped. Finally, the Davidson Boding index and average profile coefficient were used to evaluate the grouped data of multiple performance categories to obtain the sorting results of retired batteries. This significantly improved the efficiency and accuracy of retired battery sorting, meeting the needs of rapid classification, precise cascade utilization / material recycling in large-scale battery recycling, and helping to promote resource recycling and reduce industrial costs.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0011] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A schematic diagram of a data-driven method for sorting retired batteries is shown. Figure 2 A flowchart illustrating another data-driven method for sorting retired batteries is shown. Figure 3 A discharge voltage curve is shown; Figure 4 A dimensionality reduction plot of the t-SNE versus discharge voltage curve is shown; Figure 5 A grid partitioning diagram before clustering is shown; Figure 6 This shows a mesh generation diagram after applying a local gravity model; Figure 7 A classification graph using the k-means algorithm is shown. Figure 8 A discharge voltage curve after classification is shown; Figure 9 A flowchart of a data-driven rapid sorting method for retired batteries is shown. Figure 10 A schematic diagram of a data-driven sorting system for retired batteries is shown. Figure 11 A schematic diagram of another data-driven sorting structure for retired batteries is shown. Figure 12 A schematic diagram of the device structure is shown. Detailed Implementation

[0012] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0013] Currently, retired battery sorting methods are mainly divided into two categories. The first is the traditional detection method based on physical parameters, which sorts batteries by directly measuring single-point parameters such as capacity and internal resistance. However, this method relies only on local features, making it difficult to comprehensively reflect the differences in overall battery performance, resulting in limited sorting accuracy and a long detection process, which is insufficient to meet the processing needs of large-scale retired batteries. The second is the data-driven intelligent sorting method, which uses the battery's complete discharge voltage curve as a sorting feature. This method can comprehensively reflect the battery's voltage plateau, ohmic internal resistance, polarization internal resistance, and other characteristics during operation, and has become a research hotspot. Among these, the scheme combining dimensionality reduction and clustering algorithms significantly improves the potential sorting accuracy by extracting and classifying data features. However, the performance of this method is highly dependent on the sufficiency of feature extraction, the optimization of parameters such as the perplexity of t-SNE and the learning rate, and the clustering effect of the clustering algorithm. Problems such as unreasonable dimensionality reduction parameter settings and susceptibility to noise interference in traditional methods make it difficult to meet practical needs in terms of sorting efficiency and accuracy, thus hindering the large-scale and efficient recycling of retired batteries.

[0014] To address this issue, this application proposes a data-driven method for sorting retired batteries. The method selects the discharge voltage curve of lithium-ion batteries as a sorting feature, which comprehensively characterizes key performance characteristics such as voltage plateau and polarization resistance during battery operation. Compared to traditional methods relying on single-point parameters such as capacity and internal resistance, this method more completely reflects the overall performance differences of the batteries, providing more reliable basic data support for accurate sorting. The IWMA algorithm is used to optimize the perplexity, learning rate, and maximum number of iterations in t-SNE, solving the problems of poor dimensionality reduction and insufficient feature retention caused by unreasonable parameter settings in traditional t-SNE. This allows the dimensionality-reduced data to effectively compress dimensions to improve processing efficiency while preserving the key features of the original curve to the greatest extent, enhancing the accuracy of subsequent classification. This paper improves the K-means clustering algorithm by employing a grid partitioning strategy based on a local gravity model. The grid partitioning accurately reflects the data density, and a density threshold is used to distinguish between dense and non-dense regions. The introduction of a local gravity model corrects sample grid assignment, effectively eliminating noise interference and clustering only in dense regions. This solves the problems of traditional K-means being susceptible to noise and having shifted cluster centers, significantly improving the rationality and accuracy of classification. Furthermore, the Davidson Boding index (DBI) and the mean profile coefficient (SC) are used for evaluation, objectively quantifying the classification effect and ensuring the reliability of the sorting results. Through the synergistic optimization of IWMA and t-SNE, and the improved precise clustering of K-means, the efficiency and accuracy of retired battery sorting are significantly improved, effectively adapting to the processing needs of large-scale retired batteries. This provides more efficient technical support for the large-scale recycling and reuse of retired batteries, thereby improving resource recycling efficiency, reducing the safety risks of secondary applications, and possessing significant engineering application value. The implementing entity of this application may be a decommissioned battery sorting system. The decommissioned battery sorting system provides services to users by relying on the computing power of a server. The server may be an independent server or a server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0015] This application provides a data-driven method for sorting decommissioned batteries, such as... Figure 1 As shown, the method includes: 101. Obtain lithium-ion battery change data and extract the discharge voltage curve from the lithium-ion battery change data.

[0016] In this embodiment, data on the changes in lithium-ion batteries during use and aging are collected through experiments or monitoring, and a discharge voltage curve is extracted from these data. This curve can comprehensively characterize the battery's core performance characteristics, such as voltage plateau, polarization resistance, and ohmic resistance, serving as the fundamental data source for subsequent sorting. Compared to traditional methods that rely solely on single-point parameters such as capacity and internal resistance, the discharge voltage curve provides more comprehensive information, reflecting battery performance differences throughout the entire voltage change cycle and providing a reliable basis for accurate sorting.

[0017] 102. The discharge voltage curve is reduced in dimension by the optimized t-SNE algorithm to obtain the dimension-reduced voltage curve.

[0018] In this embodiment, t-SNE (t-distributed random neighborhood embedding) is a high-dimensional data dimensionality reduction algorithm; however, its performance largely depends on parameters such as perplexity, learning rate, and maximum number of iterations. The IWMA algorithm (Improved Humpback Whale Optimization) is used to intelligently optimize the key parameters of t-SNE, resulting in an optimized t-SNE algorithm that can reduce high-dimensional discharge voltage curves to low-dimensional data. This optimization solves the problems of high blindness and unstable dimensionality reduction effects of traditional t-SNE's manual parameter setting. The IWMA algorithm, through mechanisms such as population initialization, dual-mode search (global exploration and local exploitation), and adaptive perturbation, can efficiently find the optimal parameter combination for t-SNE, ensuring that the dimensionality-reduced data retains key distinguishing features of battery performance while compressing dimensionality, laying a high-quality data foundation for subsequent clustering.

[0019] 103. The improved K-means clustering algorithm was used to classify the dimensionality-reduced voltage curves, resulting in grouped data for multiple performance categories.

[0020] In this embodiment, K-means is a classic clustering algorithm; however, its traditional version is susceptible to noise interference, leading to cluster center shifts. This paper improves the K-means algorithm using a grid partitioning strategy based on a local gravity model. The specific steps are as follows: First, the dimensionality-reduced data is divided into grids, and the grid assignment of samples is corrected using a local gravity model. Next, the sample density of each grid is calculated, and dense and non-dense regions are distinguished by a density threshold. Finally, K-means clustering is performed only on samples in dense regions, resulting in multiple battery groups with different performance categories. This method eliminates noise interference in the clustering process from the source, avoiding the shortcomings of the traditional K-means algorithm, such as cluster center shifts and inaccurate category partitioning caused by noise points. The improved algorithm can accurately distinguish retired batteries with different performance characteristics, significantly improving the accuracy and reliability of clustering.

[0021] 104. Evaluate the grouped data of multiple performance categories to obtain the sorting results of retired batteries.

[0022] In this embodiment, quantitative indicators are used to evaluate multiple sets of performance data obtained from clustering, ultimately outputting the sorting results of retired batteries. The Davidson-Bourdin index (DBI) is used to measure the compactness and separation of clusters, while the mean profile coefficient (SC) is used to measure the cohesion and separation of samples. By evaluating based on objective quantitative indicators, the accuracy and rationality of the sorting results can be verified, ensuring that retired batteries of different performance categories can be accurately distinguished to meet the needs of subsequent cascade utilization (e.g., applying high-performance batteries to low-power scenarios) or material recycling (e.g., disassembling low-performance batteries to extract materials), thereby contributing to the efficient resource utilization of retired batteries.

[0023] This application provides a data-driven method for sorting retired batteries. Compared with existing technologies, this method extracts discharge voltage curves characterizing the battery's voltage plateau and polarization resistance during operation from lithium-ion battery variation data. This reflects battery performance across the entire voltage cycle, avoiding the limitations of single-point parameters and providing a reliable basis for sorting. Next, the discharge voltage curves are dimensionality-reduced using an optimized t-SNE algorithm. The optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm, solving the problem of poor dimensionality reduction caused by blindly setting parameters in traditional t-SNE. This ensures efficient compression of low-dimensional data while retaining the discriminative power of key features. Subsequently, the dimensionality-reduced voltage curves are classified using an improved K-means clustering algorithm, resulting in multiple performance category groups. The improved K-means clustering algorithm is derived by using a grid partitioning strategy based on a local gravity model, eliminating noise interference at the source, preventing cluster center shift, and ensuring accurate grouping of retired batteries with different performance characteristics. Finally, the Davidson Boding index and average profile coefficient were used to evaluate the grouped data of multiple performance categories to obtain the sorting results of retired batteries. This significantly improved the efficiency and accuracy of retired battery sorting, meeting the needs of rapid classification, precise cascade utilization / material recycling in large-scale battery recycling, and helping to promote resource recycling and reduce industrial costs.

[0024] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, and to fully illustrate the specific implementation process of this embodiment, this application provides another data-driven method for sorting retired batteries, such as... Figure 2 As shown, the method includes: 201. Obtain lithium-ion battery change data and extract the discharge voltage curve from the lithium-ion battery change data.

[0025] In this embodiment of the application, the lithium-ion battery change data is obtained by conducting battery discharge characteristic testing experiments to obtain discharge voltage data during the aging process of the lithium-ion battery. For example... Figure 3 As shown, the NASA battery dataset includes B0005, B0006, B0007, and B0018, representing different lithium-ion battery samples. The NASA battery dataset records test data of batteries during continuous charge-discharge cycles. At different cycle counts, the batteries exhibit different aging states. Therefore, batteries with different cycle counts can be considered as retired batteries with different degrees of aging, and four sets of discharge voltage data are extracted. Equating batteries with different cycle counts to retired batteries with different degrees of aging cleverly simulates various aging states of retired batteries in an experimental environment. This approach avoids the complexity of actually collecting retired batteries while covering various scenarios such as mild, moderate, and severe aging, providing diverse and representative samples for subsequent sorting algorithm verification, ensuring that the method can meet the sorting needs of retired batteries with different degrees of aging in practice.

[0026] 202. The t-SNE algorithm is optimized by using the IWMA algorithm.

[0027] In this embodiment, a normal distribution mapping is used to generate the position of each individual whale in the initial whale population. Each individual whale in the initial whale population corresponds to a set of t-SNE parameters, which include perplexity, learning rate, and maximum number of iterations, as shown in Formula 1 below: Formula 1:

[0028] in, This represents the position of the g-th individual whale in the initial whale population. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameters. denoted as a standard normally distributed random number, and g represents the index of the individual whale.

[0029] The IWMA algorithm is used to simulate the migratory behavior of humpback whales. Within each migrating pod, more experienced individuals possess higher-level location information and higher objective function values; they guide and lead the pod, directing other whales to their destination. For the current iteration number, the number of lead whales corresponding to that iteration number is determined as shown in Formula 2 below: Formula 2:

[0030] in, This indicates the number of pilot whales corresponding to the current iteration number. This represents the initial total population size of the whale population. Indicates the current iteration number. This indicates the maximum number of iterations for the IWMA algorithm. This represents the floor function. Ensure that the number of pilot whales is no less than 2.

[0031] Next, the weight of each pilot whale in the initial whale population is calculated using the number of pilot whales corresponding to the current iteration number, as shown in Formula 3 below: Formula 3:

[0032] in, This represents the weight of the i-th pilot whale in the initial whale population. This indicates the number of pilot whales corresponding to the current iteration number. denoted as exponential decay function, and i and j represent the index of the pilot whale.

[0033] Then, the weighted average position of the pilot whales corresponding to the current iteration number is calculated using the weight of each pilot whale in the initial whale population, so that non-pilot whales can move more effectively toward the position of the pilot whales, as shown in Formula 4 below: Formula 4:

[0034] in, This indicates the weighted average position of the pilot whale corresponding to the current iteration number. This indicates the number of pilot whales corresponding to the current iteration number. This represents the weight of the i-th pilot whale in the initial whale population. This represents the position of the i-th pilot whale in the initial whale population.

[0035] Subsequently, less experienced whales move towards the pilot whales' positions. By introducing a dual-mode switching mechanism, a balance between global exploration and local development is achieved. Specifically, based on the weighted average position of the pilot whales corresponding to the current iteration number, the dual-mode switching method is used to update the positions of multiple individual whales in the initial whale population, excluding the multiple pilot whales. At the same time, adaptive perturbations are applied to the multiple pilot whales in the initial whale population to update their positions, thus obtaining the whale population corresponding to the current iteration number.

[0036] The specific steps for updating the positions of multiple individual whales in the initial whale population, excluding several pilot whales, include: Whales in the initial whale population, excluding the multiple pilot whales, are designated as non-pilot whales, resulting in multiple non-pilot whales. A random number is generated for each non-pilot whale. For each non-pilot whale, the corresponding random number is then tested.

[0037] If the random number corresponding to the non-leader whale is less than 0.5, obtain the global exploration mode calculation formula, and use the global exploration mode calculation formula to update the position of the non-leader whale to obtain the updated position of the non-leader whale, as shown in Formula 5 below: Formula 5:

[0038] in, This indicates the updated position of the non-leader whale. This indicates the weighted average position of the pilot whale corresponding to the current iteration number. This represents the adaptive step size factor, which decays from 2 to 0 with each iteration. This represents a random number corresponding to a non-pilot whale. The location of the pilot whale randomly selected from the initial whale population. The location of a randomly selected whale individual within the initial whale population. represents the position of the globally optimal whale individual in the initial whale population, where l represents the index of the non-leader whale.

[0039] If the random number corresponding to the non-leader whale is greater than 0.5, obtain the local development mode calculation formula, and use the local development mode calculation formula to update the position of the non-leader whale to obtain the updated position of the non-leader whale, as shown in Formula 6 below: Formula 6:

[0040] in, This indicates the updated position of the non-leader whale. This indicates the weighted average position of the pilot whale corresponding to the current iteration number. The location is not that of the pilot whale. Levy's flight stride length This represents a random number corresponding to a non-pilot whale. represents the position of the globally optimal whale individual in the initial whale population, where l represents the index of the non-leader whale.

[0041] The adaptive perturbation update of the positions of multiple pilot whales in the initial whale population specifically includes: The pilot whale can discover and explore new areas by introducing adaptive perturbations, which broaden the search range in the early stages and improve convergence accuracy in the later stages. First, the dynamic perturbation probability corresponding to the current iteration number is calculated, as shown in Formula 7 below: Formula 7:

[0042] in, This represents the dynamic perturbation probability corresponding to the current iteration number. Indicates the current iteration number. This represents the maximum number of iterations for the IWMA algorithm. Then, a random number is generated for each pilot whale. For each pilot whale, the random number corresponding to the pilot whale is compared with the dynamic perturbation probability corresponding to the current iteration number.

[0043] If the random number corresponding to the pilot whale is less than the dynamic perturbation probability corresponding to the current iteration number, obtain the large-scale exploration calculation formula, and use the large-scale exploration calculation formula to update the position of the pilot whale, obtaining the updated position of the pilot whale, as shown in Formula 8 below: Formula 8:

[0044] in, This indicates the updated location of the pilot whale. Indicates the location of the pilot whale. This represents the random number corresponding to the pilot whale. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameters. Indicates the current iteration number. This represents the maximum number of iterations of the IWMA algorithm, and i represents the index of the pilot whale.

[0045] If the random number corresponding to the pilot whale is greater than the dynamic perturbation probability corresponding to the current iteration number, obtain the local fine search calculation formula, and use the local fine search calculation formula to update the position of the pilot whale to obtain the updated position of the pilot whale, as shown in Formula 9 below: Formula 9:

[0046] in, This indicates the updated location of the pilot whale. Indicates the location of the pilot whale. This represents the random number corresponding to the pilot whale. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameter, and i represents the index of the pilot whale.

[0047] If the current iteration count is less than the maximum iteration count of the IWMA algorithm, then the next iteration calculation is carried out based on the whale population corresponding to the current iteration count; if the current iteration count is equal to the maximum iteration count of the IWMA algorithm, then the globally optimal whale individual in the whale population corresponding to the current iteration count is used as the target parameter of the optimized t-SNE algorithm.

[0048] 203. The discharge voltage curve is reduced in dimension by the optimized t-SNE algorithm to obtain the dimension-reduced voltage curve.

[0049] In this embodiment, the discharge voltage curve of a lithium-ion battery (high-dimensional data, encompassing multi-dimensional features of voltage variation over time) is used as input. Then, the optimal parameter combination determined by the IWMA algorithm is substituted into the t-SNE algorithm to perform dimensionality reduction processing on the high-dimensional discharge voltage curve, mapping it to a set of data points in a low-dimensional space, thereby obtaining... Figure 4 The dimensionality reduction plot of the discharge voltage curve shown retains the key distinguishing features of battery performance, greatly compressing the data dimensionality while preserving the key features of the original curve that reflect the differences in battery performance to the greatest extent.

[0050] 204. The dimension-reduced voltage curve is divided into grids using a grid partitioning strategy to obtain the target grid.

[0051] In this embodiment, to accurately reflect the density of the data, the dimensionality-reduced voltage curve data is divided into a grid, and the data distribution boundary is calculated for each feature dimension to ensure that the grid covers all data points. Specifically, the data boundary for each dimension is calculated using the dimensionality-reduced voltage curve data. The data boundary includes the minimum and maximum data values, as shown in Formula 10 below: Formula 10:

[0052]

[0053] in, This represents the minimum value of the data in the m-th dimension. This represents the maximum value of the data in the m-th dimension. This represents the set of data points in the m-th dimension. This represents the floor function. This represents the floor function, where m represents the index of the dimension. Then, the interval length for each dimension is calculated using the data boundaries of each dimension, as shown in Formula 11 below: Formula 11:

[0054] in, This represents the length of the interval in the m-th dimension. This represents the minimum value of the data in the m-th dimension. This represents the maximum value of the data in the m-th dimension, where m represents the index of the dimension. Then, the number of grid cells and the size of the grid interval for each dimension are determined, as shown in Formula 12 below: Formula 12:

[0055]

[0056] in, This represents the number of grid cells in the m-th dimension. This represents the size of the grid interval in the m-th dimension, and N represents the number of samples in the dimensionality-reduced voltage curve data. This represents the length of the interval in the m-th dimension. This represents the floor function, where m represents the dimension index. Then, an initial grid is constructed based on the data boundaries, interval length, number of grids, and grid interval size for each dimension. Each sample in the dimension-reduced voltage curve data is then mapped to this initial grid to obtain the target grid, as shown in Formula 13 below: Formula 13:

[0057] in, Represents data points in the dimension-reduced voltage curve data Grid index, , This represents the nth sample in the reduced-dimensional voltage curve data. This represents the data point corresponding to the m-th dimension in the n-th sample of the dimension-reduced voltage curve data. , This represents the samples in the reduced-dimensional voltage curve data. Grid index, This represents the minimum value of the data in the m-th dimension. This represents the interval length of the m-th dimension, where n represents the index of the sample in the reduced-dimensional voltage curve data, and m represents the index of the dimension.

[0058] like Figure 5 As shown, the entire space is divided into multiple rectangular grids, each grid representing a sub-region of data. The multiple grids are further divided into dense region grid sets and non-dense region grid sets. The numbers marked inside the grids represent the number of samples in that grid. Red squares represent dense regions, where the sample points are more concentrated and the gravitational interaction between them is more pronounced. Blue squares represent non-dense regions, where the sample points are more sparsely distributed and the gravitational interaction between them is relatively weak. To optimize the grid assignment of samples, a local gravity model is introduced for correction.

[0059] 205. Based on the local gravity model, the grid affiliation of multiple samples in the target grid is corrected to obtain the corrected target grid.

[0060] In this embodiment of the application, the quality of each sample in the dimension-reduced voltage curve data is calculated as follows: Formula 14: Formula 14:

[0061] in, Let K represent the quality of the p-th sample in the dimensionality-reduced voltage curve, K represent the scaling factor, and k represent the number of nearest neighbor samples of the p-th sample in the dimensionality-reduced voltage curve. Let k represent the set of k nearest neighbors of the p-th sample. Let represent the Euclidean distance between the p-th sample and its q-th nearest neighbor sample, where p and q represent the sample indexes. Then, based on the local gravity model, the gravitational force exerted on each sample by each grid cell in the target grid is calculated using the mass of each sample and the target grid, as shown in Equation 15 below: Formula 15:

[0062] in, This indicates that the p-th sample is subject to the g-th grid. This represents the quality of the p-th sample in the dimension-reduced voltage curve. This represents the quality of the q-th sample in the reduced-dimensional voltage curve. Let represent the Euclidean distance between the p-th sample and the q-th nearest neighbor sample. Let k represent the set of k nearest neighbors of the p-th sample. Let represent the average Euclidean distance between the p-th sample and its k nearest neighbors. Let p and q represent the grid coordinates of the q-th nearest neighbor sample, and p and q represent the sample indexes in the reduced voltage curve data. Finally, a gravity matrix is ​​constructed using the gravitational force exerted on each sample by each grid in the target grid. Here, rows correspond to samples, columns correspond to grids, and elements represent the magnitude of the gravitational force exerted on the sample by the corresponding grid. Based on the gravity matrix, grid assignment correction is performed on multiple samples in the target grid, i.e., determining the grid with the strongest gravitational force on each sample, thereby correcting the sample's grid assignment and obtaining the corrected target grid.

[0063] 206. Divide multiple grids in the target grid into a set of dense grid regions and a set of non-dense grid regions.

[0064] In this embodiment, each grid in the target grid is traversed, and the number of samples in each grid is counted as the grid density of each grid. Grids with a density greater than zero are selected from the multiple grids of the target grid as non-empty grids, resulting in multiple non-empty grids. The density threshold of the target grid is determined using the grid densities of these multiple non-empty grids, as shown in Formula 16 below: Formula 16:

[0065]

[0066] in, This represents the density threshold for correcting the target mesh, where S represents an array consisting of the mesh densities of multiple non-empty meshes. Let dM represent the mesh density of the a-th non-empty mesh, dS represent the standard deviation of the mesh densities of the multiple non-empty meshes, and a represent the index of the non-empty mesh. Then, the density threshold of the modified target mesh is compared with the mesh density of each mesh to divide the multiple meshes into dense region mesh sets and non-dense region mesh sets, as shown in Formula 17 below: Formula 17:

[0067] Where C represents the set of dense region grids, and E represents the set of non-dense region grids. This represents the grid density of the b-th grid. This indicates the density threshold for the target mesh, where b represents the mesh index. For example... Figure 6 As shown, after the sample attributes are corrected by the local gravity model, the red squares represent dense areas and the blue squares represent non-dense areas.

[0068] 207. The K-means clustering algorithm was used to classify the dense region grid set to obtain data grouped into multiple performance categories.

[0069] In this embodiment, after determining the dense grid set, K-means clustering is performed on the dimensionality-reduced voltage curve samples within these grids, which have been freed from noise interference. First, the number of clusters k is preset based on the possible performance levels of the retired batteries, and k points are randomly selected from the samples as initial cluster centers. Then, the Euclidean distance between each sample and the k centers is calculated, and the sample is assigned to the category containing the nearest center. Subsequently, based on all samples within each category, the mean is recalculated as the new cluster center. This iterative process of distance calculation, sample assignment, and center update is repeated until the cluster center positions stabilize, i.e., the change is less than a preset threshold. The resulting k categories represent multiple performance category groupings, where batteries within each group have similar discharge voltage curve characteristics, corresponding to similar aging degrees and performance levels. Figure 7 As shown, the X and Y axes are the battery feature dimensions after t-SNE dimensionality reduction, representing the key low-dimensional features of the discharge voltage curve; different colored points correspond to retired batteries of different performance categories, with red representing category 1, green representing category 2, and blue representing category 3, and black dots being the centers of each cluster.

[0070] 208. Evaluate the grouped data of multiple performance categories to obtain the sorting results of retired batteries.

[0071] In this embodiment of the application, for each performance category grouping data, the Davidsonburg index of the performance category grouping data is calculated as shown in the following formula 18: Formula 18:

[0072]

[0073] in, The Davidson-Bolding index represents the performance category grouping data, where r represents the number of performance category groupings. This represents the cluster divergence of the data grouped by the c-th performance category. This represents the cluster divergence of the data grouped by the o-th performance category. This represents the number of samples in the data group for the o-th performance category. This represents the cluster center of the data grouped by the 0th performance category. This represents the cluster center of the c-th performance category grouping data. Let f represent the distance between the cluster center of the c-th performance category group and the cluster center of the o-th performance category group, and let f represent the f-th sample of the o-th performance category group. This represents the average distance from the f-th sample in the o-th performance category group to all other samples in the o-th performance category group except for the f-th sample, where o and c represent the index of the performance category group, and f represents the index of the sample in the performance category group.

[0074] Next, the average silhouette coefficient of the performance category grouped data is calculated as shown in Formula 19 below: Formula 19:

[0075]

[0076] in, The mean silhouette coefficient represents the performance category grouping data, and T represents the number of samples in the performance category grouping data. This represents the silhouette coefficient of the t-th sample in the performance category grouping data. This represents the separation of the t-th sample in the performance category grouping data. This represents the cohesion of the t-th sample in the performance category grouping data, where t represents the index of the sample in the performance category grouping data.

[0077] The Davidson Boding index and average profile coefficient of each performance category group are used as the evaluation results for each performance category group. The evaluation results of multiple performance category group data are used as the sorting results for retired batteries. Figure 8As shown, the horizontal axis represents discharge time, and the vertical axis represents battery voltage. Cluster 1, Cluster 2, and Cluster 3 represent three groups of retired batteries with different performance obtained through improved K-Means clustering. Cluster 3 has the highest voltage plateau, the longest duration, and the slowest voltage drop rate in the later stage, representing retired batteries with better performance and lower aging. Cluster 1 has the lowest voltage plateau and the fastest voltage drop in the later stage, representing retired batteries with higher aging and more obvious performance degradation. The performance of Cluster 2 is between the two.

[0078] In summary, such as Figure 9The flowchart shown illustrates a data-driven rapid sorting method for retired batteries. Through battery discharge characteristic testing experiments, discharge voltage data of lithium-ion batteries during the aging process is collected, and a discharge voltage curve is extracted. This curve comprehensively characterizes the battery's core performance features, such as voltage plateau, polarization resistance, and ohmic resistance, providing multi-dimensional, full-cycle foundational data for subsequent sorting and avoiding the limitations of traditional single-point parameters (such as capacity and internal resistance). For high-dimensional discharge voltage curves, the Improved Humpback Whale Algorithm (IWMA) is used to globally optimize key parameters (perplexity, learning rate, and maximum number of iterations) of the t-SNE algorithm. The specific process includes: initializing individual algorithm positions based on a normal distribution; dynamically generating the number of pilot whales and calculating their weighted average position; non-pilot whales performing a dual-mode optimization of global exploration (introducing random positions to expand the search range) + local development (fine-grained search based on Levy flight step size); and pilot whales dynamically adjusting their search strategy through adaptive perturbations. After parameter optimization, the optimal parameters are substituted into the t-SNE algorithm to reduce the dimensionality of the high-dimensional voltage curve, resulting in low-dimensional data that retains key distinguishing features of battery performance. For the dimensionality-reduced data, a local gravity model-based grid partitioning strategy is used to improve K-means clustering: first, grid partitioning is performed, and the sample density of each grid is calculated. A density threshold is used to distinguish dense regions (effective samples) from non-dense regions (noise); then, the grid affiliation of samples is corrected using a local gravity model (combining sample quality, Euclidean distance, and distance decay factor), and k-means++ clustering is performed only on samples in dense regions. After clustering, the Davidson-Bolding index (DBI) and mean profile coefficient (SC) are calculated to evaluate the sorting effect. If the threshold condition of "DBI < 0.85 and SC > 0.35" is met, the performance category grouping results of the retired batteries are output; otherwise, the parameters are adjusted, and the dimensionality reduction, clustering, and evaluation process is repeated until the required sorting results are obtained. This application employs a sorting method combining t-SNE without parameter optimization with traditional K-means, and compares it with the method of this invention, which optimizes t-SNE using the IWMA algorithm and improves K-means by incorporating a local gravity model. The method of this invention reduces the DBI by 12.3% and improves the average profile coefficient (SC) by 15.1% compared to the traditional method; this indicates that the data-driven rapid sorting method for decommissioned batteries proposed in this invention has higher sorting accuracy.

[0079] This application provides a data-driven method for sorting retired batteries. Compared with existing technologies, this method extracts discharge voltage curves characterizing the battery's voltage plateau and polarization resistance during operation from lithium-ion battery variation data. This reflects battery performance across the entire voltage cycle, avoiding the limitations of single-point parameters and providing a reliable basis for sorting. Next, the discharge voltage curves are dimensionality-reduced using an optimized t-SNE algorithm. The optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm, solving the problem of poor dimensionality reduction caused by blindly setting parameters in traditional t-SNE. This ensures efficient compression of low-dimensional data while retaining the discriminative power of key features. Subsequently, the dimensionality-reduced voltage curves are classified using an improved K-means clustering algorithm, resulting in multiple performance category groups. The improved K-means clustering algorithm is derived by using a grid partitioning strategy based on a local gravity model, eliminating noise interference at the source, preventing cluster center shift, and ensuring accurate grouping of retired batteries with different performance characteristics. Finally, the Davidson Boding index and average profile coefficient were used to evaluate the grouped data of multiple performance categories to obtain the sorting results of retired batteries. This significantly improved the efficiency and accuracy of retired battery sorting, meeting the needs of rapid classification, precise cascade utilization / material recycling in large-scale battery recycling, and helping to promote resource recycling and reduce industrial costs.

[0080] Furthermore, as Figure 1 In a specific implementation of the method, this application provides a data-driven decommissioned battery sorting device, such as... Figure 10 As shown, the device includes: an acquisition module 301, a dimensionality reduction module 302, a classification module 303, and an evaluation module 304.

[0081] The acquisition module 301 is used to acquire lithium-ion battery change data and extract the discharge voltage curve from the lithium-ion battery change data. Dimensionality reduction module 302 is used to perform dimension reduction processing on the discharge voltage curve using the optimized t-SNE algorithm to obtain the dimension-reduced voltage curve. The optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm. The classification module 303 is used to classify the dimension-reduced voltage curves using an improved K-means clustering algorithm to obtain multiple performance category grouping data. The improved K-means clustering algorithm is obtained by improving the K-means clustering algorithm using the grid partitioning strategy of the local gravity model. Evaluation module 304 is used to evaluate the grouped data of the multiple performance categories to obtain the sorting results of retired batteries.

[0082] In specific application scenarios, such as Figure 11 As shown, the device also includes a t-SNE algorithm optimization module 305.

[0083] The t-SNE algorithm optimization module 305 is used to generate the position of each individual whale in the initial whale population using a normal distribution mapping. Each individual whale in the initial whale population corresponds to a set of t-SNE parameters, including perplexity, learning rate, and maximum number of iterations.

[0084] in, This indicates the position of the g-th individual whale in the initial whale population. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameters. Let g represent a standard normally distributed random number, and g represent the index of an individual whale. For the current iteration number, determine the number of pilot whales corresponding to the current iteration number.

[0085] in, This indicates the number of pilot whales corresponding to the current iteration number. This represents the total population size of the initial whale population. This indicates the current iteration number. This represents the maximum number of iterations for the IWMA algorithm. This represents the floor function; the weight of each pilot whale in the initial whale population is calculated using the number of pilot whales corresponding to the current iteration number.

[0086] in, This represents the weight of the i-th pilot whale in the initial whale population. This represents the number of pilot whales corresponding to the current iteration number, where i and j represent the index of the pilot whales. The weighted average position of the pilot whales corresponding to the current iteration number is calculated using the weights corresponding to each pilot whale in the initial whale population.

[0087] in, This indicates the weighted average position of the pilot whale corresponding to the current iteration number. This indicates the number of pilot whales corresponding to the current iteration number. This represents the weight of the i-th pilot whale in the initial whale population. Let represent the position of the i-th pilot whale in the initial whale population. Based on the weighted average position of the pilot whales corresponding to the current iteration number, the positions of multiple whale individuals in the initial whale population (excluding multiple pilot whales) are updated using a dual-mode switching method, and the positions of multiple pilot whales in the initial whale population are updated with adaptive perturbation to obtain the whale population corresponding to the current iteration number. If the current iteration number is less than the maximum iteration number of the IWMA algorithm, the next iteration calculation is performed based on the whale population corresponding to the current iteration number. If the current iteration number is equal to the maximum iteration number of the IWMA algorithm, the globally optimal whale individual in the whale population corresponding to the current iteration number is used as the target parameter of the optimized t-SNE algorithm.

[0088] In a specific application scenario, the t-SNE algorithm optimization module 305 is used to classify all whale individuals in the initial whale population, excluding multiple pilot whales, as non-pilot whales, thus obtaining multiple non-pilot whales; generate a random number for each non-pilot whale; for each non-pilot whale, check the random number corresponding to the non-pilot whale; if the random number corresponding to the non-pilot whale is less than 0.5, obtain the global exploration mode calculation formula, and use the global exploration mode calculation formula to update the position of the non-pilot whales, thus obtaining the updated position of the non-pilot whales.

[0089] in, This indicates the updated position of the non-pilot whale. This indicates the weighted average position of the pilot whale corresponding to the current iteration number. Indicates the adaptive step size factor. This represents the random number corresponding to the non-pilot whale. The location of the pilot whale randomly selected from the initial whale population. The location of a randomly selected whale individual within the initial whale population. Let l represent the position of the globally optimal whale individual in the initial whale population, and l represent the index of the non-leader whale. If the random number corresponding to the non-leader whale is greater than 0.5, the local development mode calculation formula is obtained, and the position of the non-leader whale is updated using the local development mode calculation formula to obtain the updated position of the non-leader whale.

[0090] in, This indicates the updated position of the non-pilot whale. This indicates the weighted average position of the pilot whale corresponding to the current iteration number. The location of the non-pilot whale. Levy's flight stride length This represents the random number corresponding to the non-pilot whale. Let l represent the position of the globally optimal whale individual in the initial whale population, where l represents the index of the non-leader whale.

[0091] In specific application scenarios, the t-SNE algorithm optimization module 305 is used to calculate the dynamic perturbation probability corresponding to the current iteration number.

[0092] in, This represents the dynamic perturbation probability corresponding to the current iteration number. This indicates the current iteration number. The maximum number of iterations for the IWMA algorithm is represented; a random number is generated for each pilot whale; for each pilot whale, the random number corresponding to the pilot whale is compared with the dynamic perturbation probability corresponding to the current iteration number; if the random number corresponding to the pilot whale is less than the dynamic perturbation probability corresponding to the current iteration number, a large-scale exploration calculation formula is obtained, and the pilot whale's position is updated using the large-scale exploration calculation formula to obtain the updated position of the pilot whale.

[0093] in, This indicates the updated position of the pilot whale. This indicates the location of the pilot whale. This represents the random number corresponding to the pilot whale. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameters. This indicates the current iteration number. Let represent the maximum number of iterations of the IWMA algorithm, and let i represent the index of the pilot whale. If the random number corresponding to the pilot whale is greater than the dynamic perturbation probability corresponding to the current iteration number, the local fine search calculation formula is obtained, and the pilot whale's position is updated using the local fine search calculation formula to obtain the updated position of the pilot whale.

[0094] in, This indicates the updated position of the pilot whale. This indicates the location of the pilot whale. This represents the random number corresponding to the pilot whale. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameter, and i represents the index of the pilot whale.

[0095] In specific application scenarios, the classification module 303 is used to perform grid partitioning on the dimension-reduced voltage curve using the aforementioned grid partitioning strategy to obtain the target grid; and to calculate the quality of each sample in the dimension-reduced voltage curve data.

[0096] in, The quality of the p-th sample in the reduced-dimensional voltage curve is represented by K, where K represents the scaling factor and k represents the number of nearest neighbor samples of the p-th sample in the reduced-dimensional voltage curve. Let k represent the set of k nearest neighbors of the p-th sample. Let represent the Euclidean distance between the p-th sample and its q-th nearest neighbor sample, where p and q represent the sample indexes; based on the local gravity model, the gravitational force exerted on each sample by each grid cell in the target grid is calculated using the mass of each sample and the target grid.

[0097] in, This indicates that the p-th sample is subject to the g-th grid. This represents the quality of the p-th sample in the reduced-dimensional voltage curve. This represents the quality of the q-th sample in the reduced-dimensional voltage curve. Let represent the Euclidean distance between the p-th sample and the q-th nearest neighbor sample. Let k represent the set of k nearest neighbors of the p-th sample. Let represent the average Euclidean distance between the p-th sample and its k nearest neighbors. Let p and q represent the grid coordinates of the q-th nearest neighbor sample, and p and q represent the index of the sample in the reduced voltage curve data. A gravity matrix is ​​constructed using the gravitational force exerted on each sample by each grid in the target grid. Based on this gravity matrix, grid affiliation correction is performed on multiple samples in the target grid to obtain a corrected target grid. Each grid in the corrected target grid is traversed, and the number of samples in each grid is counted as the grid density of each grid. Grids with a density greater than zero are selected from the multiple grids of the corrected target grid as non-empty grids, resulting in multiple non-empty grids. The density threshold of the corrected target grid is determined using the grid densities of these multiple non-empty grids.

[0098]

[0099] in, This represents the density threshold of the target mesh, and S represents an array consisting of the mesh densities of the plurality of non-empty meshes. Let dM represent the grid density of the a-th non-empty grid, dS represent the standard deviation of the grid densities of the multiple non-empty grids, and a represent the index of the non-empty grid. The density threshold of the modified target grid is compared with the grid density of each grid to divide the multiple grids into a dense region grid set and a non-dense region grid set.

[0100] Where C represents the set of dense region grids, and E represents the set of non-dense region grids. This represents the grid density of the b-th grid. denoted by , b represents the density threshold of the target grid for correction, and b represents the grid index; the K-means clustering algorithm is used to classify the dense region grid set to obtain the grouped data of the multiple performance categories.

[0101] In specific application scenarios, the classification module 303 is used to calculate the data boundary for each dimension using the dimensionality-reduced voltage curve data. The data boundary includes the minimum and maximum data values.

[0102]

[0103] in, This represents the minimum value of the data in the m-th dimension. This represents the maximum value of the data in the m-th dimension. This represents the set of data points in the m-th dimension. This represents the floor function. This represents the floor function, where m represents the index of the dimension; the interval length of each dimension is calculated using the data boundaries of each dimension.

[0104] in, This represents the length of the interval in the m-th dimension. This represents the minimum value of the data in the m-th dimension. This represents the maximum data value in the m-th dimension, where m represents the index of the dimension; the number of grid cells and the size of the grid interval are determined for each dimension.

[0105]

[0106] in, This represents the number of grid cells in the m-th dimension. This represents the size of the grid interval in the m-th dimension, and N represents the number of samples in the dimensionality-reduced voltage curve data. This represents the length of the interval in the m-th dimension. The function represents the floor function, and m represents the index of the dimension. An initial grid is constructed based on the data boundaries, interval length, number of grids, and grid interval size for each dimension. Each sample in the dimension-reduced voltage curve data is mapped to the initial grid to obtain the target grid.

[0107] in, This represents the data points in the reduced-dimensional voltage curve data. Grid index, , This represents the nth sample in the reduced-dimensional voltage curve data. This represents the data point corresponding to the m-th dimension in the n-th sample of the dimensionality-reduced voltage curve data. , This indicates the samples in the reduced voltage curve data. Grid index, This represents the minimum value of the data in the m-th dimension. This represents the interval length of the m-th dimension, where n represents the index of the sample in the reduced-dimensional voltage curve data, and m represents the index of the dimension.

[0108] In specific application scenarios, the evaluation module 304 is used to calculate the Davidsonburg index for each performance category group of data.

[0109]

[0110] in, The Davidson-Boldt index represents the performance category grouping data, and r represents the number of performance category grouping data. This represents the cluster divergence of the data grouped by the c-th performance category. This represents the cluster divergence of the data grouped by the o-th performance category. This represents the number of samples in the data group for the o-th performance category. This represents the cluster center of the data grouped by the 0th performance category. This represents the cluster center of the c-th performance category grouping data. Let f represent the distance between the cluster center of the c-th performance category group and the cluster center of the o-th performance category group, and let f represent the f-th sample of the o-th performance category group. Let represent the average distance from the f-th sample in the o-th performance category group to all other samples in the o-th performance category group except for the f-th sample, where o and c represent the index of the performance category group, and f represents the index of the sample in the performance category group; calculate the average silhouette coefficient of the performance category group data.

[0111]

[0112] in, The mean silhouette coefficient represents the performance category grouping data, and T represents the number of samples in the performance category grouping data. This represents the silhouette coefficient of the t-th sample in the performance category grouping data. This represents the separation degree of the t-th sample in the performance category grouping data. The cohesion of the t-th sample in the performance category grouping data is represented by t, where t represents the index of the sample in the performance category grouping data. The Davidson Boding index and the average profile coefficient of each performance category grouping data are used as the evaluation results of each performance category grouping data, and the evaluation results of the multiple performance category grouping data are used as the sorting results of the retired batteries.

[0113] This application provides a data-driven retired battery sorting device. Compared with existing technologies, this application extracts discharge voltage curves characterizing the battery's voltage plateau, polarization resistance, and other properties during operation from lithium-ion battery variation data. This reflects battery performance across the entire voltage cycle, avoiding the limitations of single-point parameters and providing a reliable basis for sorting. Next, the discharge voltage curves are dimensionality-reduced using an optimized t-SNE algorithm. The optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm, solving the problem of poor dimensionality reduction caused by blindly setting parameters in traditional t-SNE. This ensures efficient compression of low-dimensional data while retaining the discriminative power of key features. Subsequently, the dimensionality-reduced voltage curves are classified using an improved K-means clustering algorithm, resulting in multiple performance category groups. The improved K-means clustering algorithm is obtained by improving the K-means clustering algorithm using a grid partitioning strategy based on a local gravity model. This eliminates noise interference at the source, prevents cluster center shift, and allows retired batteries with different performance characteristics to be accurately grouped. Finally, the Davidson Boding index and average profile coefficient were used to evaluate the grouped data of multiple performance categories to obtain the sorting results of retired batteries. This significantly improved the efficiency and accuracy of retired battery sorting, meeting the needs of rapid classification, precise cascade utilization / material recycling in large-scale battery recycling, and helping to promote resource recycling and reduce industrial costs.

[0114] It should be noted that other corresponding descriptions of the functional units involved in the data-driven decommissioned battery sorting device provided in this application embodiment can be found in the following references. Figure 1 and Figures 2 to 9 The corresponding descriptions in [the document] will not be repeated here.

[0115] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0116] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0117] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

[0118] In an exemplary embodiment, see Figure 12 The invention also provides a device comprising a bus, a processor, a memory, and a communication interface, and may further include an input / output interface and a display device, wherein the various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the data-driven decommissioned battery sorting method described in the above embodiments.

[0119] A medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the data-driven decommissioned battery sorting method described above.

[0120] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented in hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0121] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application.

[0122] Those skilled in the art will understand that the modules in the apparatus of the implementation scenario can be distributed within the apparatus of the implementation scenario as described, or they can be located in one or more apparatuses different from this implementation scenario, with corresponding changes. The modules of the above-described implementation scenario can be combined into one module, or they can be further divided into multiple sub-modules.

[0123] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of the implementation scenario.

[0124] The above disclosures are only a few specific implementation scenarios of this application. However, this application is not limited to these. Any variations that can be conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A data-driven method for sorting decommissioned batteries, characterized in that, include: Acquire lithium-ion battery variation data and extract the discharge voltage curve from the lithium-ion battery variation data; The discharge voltage curve is reduced in dimension by the optimized t-SNE algorithm to obtain the dimension-reduced voltage curve. The optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm. The reduced voltage curves are classified using an improved K-means clustering algorithm to obtain multiple performance category groupings. The improved K-means clustering algorithm is obtained by improving the K-means clustering algorithm using the grid partitioning strategy of the local gravity model. The data grouped by the multiple performance categories are evaluated to obtain the sorting results of retired batteries.

2. The method according to claim 1, characterized in that, The method further includes: The location of each individual whale in the initial whale population is generated using a normal distribution mapping. Each individual whale in the initial whale population corresponds to a set of t-SNE parameters, which include perplexity, learning rate, and maximum number of iterations. in, This indicates the position of the g-th individual whale in the initial whale population. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameters. denoted as a standard normally distributed random number, g represents the index of the individual whale; For the current iteration number, determine the number of pilot whales corresponding to the current iteration number. in, This indicates the number of pilot whales corresponding to the current iteration number. This represents the total population size of the initial whale population. This indicates the current iteration number. This represents the maximum number of iterations for the IWMA algorithm. Represents the floor function; The weight of each pilot whale in the initial whale population is calculated using the number of pilot whales corresponding to the current iteration number. in, This represents the weight of the i-th pilot whale in the initial whale population. This indicates the number of pilot whales corresponding to the current iteration number, where i and j represent the index of the pilot whale. Calculate the weighted average position of the pilot whales corresponding to the current iteration number using the weights corresponding to each pilot whale in the initial whale population. in, This indicates the weighted average position of the pilot whale corresponding to the current iteration number. This indicates the number of pilot whales corresponding to the current iteration number. This represents the weight of the i-th pilot whale in the initial whale population. This indicates the position of the i-th pilot whale in the initial whale population; Based on the weighted average position of the pilot whales corresponding to the current iteration number, the positions of multiple whale individuals in the initial whale population (excluding multiple pilot whales) are updated using a dual-mode switching method, and the positions of multiple pilot whales in the initial whale population are updated by adaptive perturbation, thus obtaining the whale population corresponding to the current iteration number. If the current iteration number is less than the maximum iteration number of the IWMA algorithm, then the next iteration calculation is performed based on the whale population corresponding to the current iteration number; If the current iteration number is equal to the maximum iteration number of the IWMA algorithm, then the globally optimal whale individual in the whale population corresponding to the current iteration number is taken as the target parameter of the optimized t-SNE algorithm.

3. The method according to claim 2, characterized in that, The step of updating the positions of multiple whale individuals in the initial whale population (excluding multiple pilot whales) based on the weighted average position of the pilot whales corresponding to the current iteration number, using a dual-mode switching approach, includes: The whales in the initial whale population other than the multiple pilot whales were considered as non-pilot whales, resulting in multiple non-pilot whales; Generate a random number for each of the non-leader whales; For each non-leader whale, the random number corresponding to the non-leader whale is detected; If the random number corresponding to the non-leader whale is less than 0.5, the global exploration mode calculation formula is obtained, and the position of the non-leader whale is updated using the global exploration mode calculation formula to obtain the updated position of the non-leader whale. in, This indicates the updated position of the non-pilot whale. This indicates the weighted average position of the pilot whale corresponding to the current iteration number. Indicates the adaptive step size factor. This represents the random number corresponding to the non-pilot whale. The location of the pilot whale randomly selected from the initial whale population. The location of a randomly selected whale individual within the initial whale population. The position of the globally optimal whale individual in the initial whale population, where l represents the index of the non-leader whale; If the random number corresponding to the non-leader whale is greater than 0.5, the local development mode calculation formula is obtained, and the position of the non-leader whale is updated using the local development mode calculation formula to obtain the updated position of the non-leader whale. in, This indicates the updated position of the non-pilot whale. This indicates the weighted average position of the pilot whale corresponding to the current iteration number. The location of the non-pilot whale. Levy's flight stride length This represents the random number corresponding to the non-pilot whale. Let l represent the position of the globally optimal whale individual in the initial whale population, where l represents the index of the non-leader whale.

4. The method according to claim 2, characterized in that, The adaptive perturbation update of the positions of multiple pilot whales in the initial whale population includes: Calculate the dynamic perturbation probability corresponding to the current iteration number. in, This represents the dynamic perturbation probability corresponding to the current iteration number. This indicates the current iteration number. This indicates the maximum number of iterations for the IWMA algorithm. Generate a random number for each of the pilot whales; For each pilot whale, the random number corresponding to the pilot whale is compared with the dynamic perturbation probability corresponding to the current iteration number; If the random number corresponding to the pilot whale is less than the dynamic perturbation probability corresponding to the current iteration number, a large-scale exploration calculation formula is obtained, and the pilot whale's position is updated using the large-scale exploration calculation formula to obtain the updated position of the pilot whale. in, This indicates the updated position of the pilot whale. This indicates the location of the pilot whale. This represents the random number corresponding to the pilot whale. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameters. This indicates the current iteration number. This represents the maximum number of iterations of the IWMA algorithm, where i represents the index of the pilot whale. If the random number corresponding to the pilot whale is greater than the dynamic perturbation probability corresponding to the current iteration number, the local fine-search calculation formula is obtained, and the pilot whale's position is updated using the local fine-search calculation formula to obtain the updated position of the pilot whale. in, This indicates the updated position of the pilot whale. This indicates the location of the pilot whale. This represents the random number corresponding to the pilot whale. Indicates the lower bound of the t-SNE parameter. This indicates the upper bound of the t-SNE parameter, and i represents the index of the pilot whale.

5. The method according to claim 1, characterized in that, The improved K-means clustering algorithm is used to classify the dimensionality-reduced voltage curves, resulting in multiple performance category groups, including: The grid partitioning strategy described above is used to partition the dimension-reduced voltage curve into a target grid. Calculate the quality of each sample in the dimensionality-reduced voltage curve data. in, The quality of the p-th sample in the reduced-dimensional voltage curve is represented by K, where K represents the scaling factor and k represents the number of nearest neighbor samples of the p-th sample in the reduced-dimensional voltage curve. Let k represent the set of k nearest neighbors of the p-th sample. Let represent the Euclidean distance between the p-th sample and the q-th nearest neighbor sample, where p and q represent the sample index. Based on the local gravity model, the gravitational force exerted on each sample by each grid cell in the target grid is calculated using the mass of each sample and the target grid. in, This indicates that the p-th sample is subject to the g-th grid. This represents the quality of the p-th sample in the reduced-dimensional voltage curve. This represents the quality of the q-th sample in the reduced-dimensional voltage curve. Let represent the Euclidean distance between the p-th sample and the q-th nearest neighbor sample. Let k represent the set of k nearest neighbors of the p-th sample. Let represent the average Euclidean distance between the p-th sample and its k nearest neighbors. The grid coordinates of the q-th nearest neighbor sample are represented by p and q, where p and q represent the index of the sample in the reduced voltage curve data. A gravity matrix is ​​constructed using the gravitational force exerted on each sample by each grid in the target grid. Based on the gravity matrix, the grid affiliation of multiple samples in the target grid is corrected to obtain the corrected target grid. Traverse each grid in the target grid and count the number of samples in each grid, which is used as the grid density of each grid; Among the multiple grids of the target grid for correction, grids with a density greater than zero are selected as non-empty grids, resulting in multiple non-empty grids. The density threshold of the target grid for correction is determined using the grid density of these multiple non-empty grids. in, This represents the density threshold of the target mesh, and S represents an array consisting of the mesh densities of the plurality of non-empty meshes. dM represents the grid density of the a-th non-empty grid, dS represents the standard deviation of the grid density of the multiple non-empty grids, and a represents the index of the non-empty grid. The density threshold of the modified target mesh is compared with the mesh density of each mesh to divide the multiple meshes into a dense region mesh set and a non-dense region mesh set. Where C represents the set of dense region grids, and E represents the set of non-dense region grids. This represents the grid density of the b-th grid. represents the density threshold of the target mesh being corrected, and b represents the mesh index; The K-means clustering algorithm is used to classify the dense region grid set to obtain the grouped data of the multiple performance categories.

6. The method according to claim 1, characterized in that, The step of using the aforementioned mesh partitioning strategy to partition the dimension-reduced voltage curve into a target mesh includes: The data boundary for each dimension is calculated using the reduced voltage curve data, and the data boundary includes the minimum and maximum data values. in, This represents the minimum value of the data in the m-th dimension. This represents the maximum value of the data in the m-th dimension. This represents the set of data points in the m-th dimension. This represents the floor function. This represents the floor function, where m represents the index of the dimension. Calculate the interval length for each dimension using the data boundaries of each dimension. in, This represents the length of the interval in the m-th dimension. This represents the minimum value of the data in the m-th dimension. This represents the maximum value of the data in the m-th dimension, where m represents the index of the dimension. Determine the number of grid cells and the size of the grid interval for each dimension. in, This represents the number of grid cells in the m-th dimension. This represents the size of the grid interval in the m-th dimension, and N represents the number of samples in the dimensionality-reduced voltage curve data. This represents the length of the interval in the m-th dimension. This represents the floor function, where m represents the index of the dimension. An initial grid is constructed based on the data boundaries, interval length, number of grids, and grid interval size for each dimension. Each sample in the dimensionality-reduced voltage curve data is then mapped to this initial grid to obtain the target grid. in, This represents the data points in the reduced-dimensional voltage curve data. Grid index, , This represents the nth sample in the reduced-dimensional voltage curve data. This represents the data point corresponding to the m-th dimension in the n-th sample of the dimensionality-reduced voltage curve data. , This indicates the samples in the reduced voltage curve data. Grid index, This represents the minimum value of the data in the m-th dimension. This represents the interval length of the m-th dimension, where n represents the index of the sample in the reduced-dimensional voltage curve data, and m represents the index of the dimension.

7. The method according to claim 1, characterized in that, The evaluation of the grouped data of the multiple performance categories to obtain the sorting results of retired batteries includes: For each performance category group, calculate the Davidsonburg index for that performance category group. in, The Davidson-Boldt index represents the performance category grouping data, and r represents the number of performance category grouping data. This represents the cluster divergence of the data grouped by the c-th performance category. This represents the cluster divergence of the data grouped by the o-th performance category. This represents the number of samples in the data group for the o-th performance category. This represents the cluster center of the data grouped by the 0th performance category. This represents the cluster center of the c-th performance category grouping data. Let f represent the distance between the cluster center of the c-th performance category group and the cluster center of the o-th performance category group, and let f represent the f-th sample of the o-th performance category group. This represents the average distance between the f-th sample in the o-th performance category group and all other samples in the o-th performance category group except for the f-th sample, where o and c represent the index of the performance category group, and f represents the index of the sample in the performance category group. Calculate the average silhouette coefficient of the performance category grouped data. in, The mean silhouette coefficient represents the performance category grouping data, and T represents the number of samples in the performance category grouping data. This represents the silhouette coefficient of the t-th sample in the performance category grouping data. This represents the separation degree of the t-th sample in the performance category grouping data. This represents the cohesion of the t-th sample in the performance category grouping data, where t represents the index of the sample in the performance category grouping data; The Davidson Boding index and average profile coefficient of each performance category group data are used as the evaluation result of each performance category group data, and the evaluation results of the multiple performance category group data are used as the sorting result of the retired batteries.

8. A data-driven retired battery sorting device, applied to the data-driven retired battery sorting method of claim 1, characterized in that, include: The acquisition module is used to acquire lithium-ion battery change data and extract the discharge voltage curve from the lithium-ion battery change data. The dimension reduction module is used to perform dimension reduction processing on the discharge voltage curve using the optimized t-SNE algorithm to obtain the dimension-reduced voltage curve. The optimized t-SNE algorithm is obtained by optimizing the t-SNE algorithm using the IWMA algorithm. The classification module is used to classify the dimension-reduced voltage curves using an improved K-means clustering algorithm to obtain multiple performance category groupings. The improved K-means clustering algorithm is obtained by improving the K-means clustering algorithm using the grid partitioning strategy of the local gravity model. The evaluation module is used to evaluate the grouped data of the multiple performance categories to obtain the sorting results of retired batteries.

9. A device comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A 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 according to any one of claims 1 to 7.

Images