A Smart Sorting Method and System for Retired Lithium Batteries Based on Dynamic Evolution of Aging Relationship Diagram

By using a method based on the dynamic evolution of aging relationship graphs to dynamically update the graph topology of retired battery groups, the problem of insufficient description of aging relationships in retired battery groups is solved, and efficient and low-cost intelligent sorting and comprehensive utilization of retired batteries is achieved.

CN123115689APending Publication Date: 2026-11-17安徽国麒科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611425569.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies are difficult to effectively describe the aging relationships of retired lithium battery groups. Furthermore, existing methods typically train retired batteries as independent samples, ignoring the aging correlations within the group. This results in poor generalization performance of the model across manufacturers, batches, and service conditions, and high annotation costs, making it difficult to meet the needs of rapid detection.

Method used

An aging relationship graph-based dynamic evolution method is adopted. An aging relationship graph is constructed using multi-source detection data. The graph topology is dynamically updated by utilizing a two-way collaborative optimization mechanism of graph structure and node representation to achieve intelligent sorting of retired battery groups.

Benefits of technology

It improves the ability to express the health status of retired battery populations, reduces label dependence and detection costs, enhances the generalization ability of the model, and is suitable for rapid sorting and comprehensive utilization of large-scale retired batteries.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115689A_ABST
    Figure CN123115689A_ABST
Patent Text Reader

Abstract

This invention proposes an intelligent sorting method and system for retired lithium batteries based on dynamic evolution of aging relationship graphs. The method includes collecting multi-source detection data from retired lithium batteries and generating a node feature matrix after processing. A first aging correlation degree and a first aging relationship graph are obtained based on the aging similarity and weight vectors between each retired lithium battery node. The node feature matrix and the initialized aging relationship graph are used as input to a graph coding network, and the deep aging features of all retired lithium battery nodes are output. Through iterative updates, a second aging correlation degree and a second aging relationship graph are generated. If the second aging relationship graph satisfies the graph structure convergence criterion, a third aging relationship graph is generated. Intelligent sorting of retired lithium batteries is completed based on the third aging relationship graph. This invention enables the relationship network between retired batteries to continuously evolve with model training, gradually approximating the real aging relationship network and improving the ability to express the health status of complex retired battery groups.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of comprehensive utilization of retired power batteries and artificial intelligence technology, and involves graph representation learning, dynamic graph learning and health status assessment technology of retired lithium-ion batteries. Specifically, it involves a method and system for intelligent sorting of retired lithium batteries based on aging relationship graph evolution (ARGE). Background Technology

[0002] Currently, the comprehensive utilization of retired lithium-ion batteries mainly relies on capacity testing, AC internal resistance testing, electrochemical impedance spectroscopy (EIS) analysis, and manual judgment to assess their health status. Among these methods, capacity testing requires a complete charge-discharge cycle, typically taking several hours, making it difficult to meet the needs of large-scale, rapid testing of retired batteries. While AC internal resistance testing is relatively fast, it only reflects the degree of battery polarization and cannot accurately describe the complex aging state inside the battery. Electrochemical impedance spectroscopy can accurately characterize the internal electrochemical properties of the battery, but the testing equipment is expensive and the testing process is complex, making it unsuitable for the rapid sorting of large-scale retired batteries.

[0003] In recent years, with the development of artificial intelligence technology, methods such as Support Vector Machine (SVM), Random Forest, XGBoost, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Transformer have been gradually applied to the prediction of the health status and comprehensive utilization sorting of retired batteries. These methods typically treat each battery as an independent sample, using detection data to establish a mapping relationship between input features and the state of health (SOH) or capacity, which improves prediction accuracy to some extent. However, these methods generally rely on a large number of labeled samples, while the true SOH of retired batteries usually needs to be obtained through complete capacity testing, which is costly and time-consuming, making the models difficult to apply to real-world engineering scenarios. In addition, the aging patterns of retired batteries from different manufacturers, with different material systems, and under different service conditions vary significantly, making it difficult for the same model to generalize across batches of data.

[0004] In recent years, self-supervised learning and graph representation learning have been increasingly applied in the field of industrial intelligent inspection. Self-supervised learning can learn sample feature representations using unlabeled data, while graph neural networks can improve model performance by utilizing the relationships between samples, providing new research ideas for the intelligent sorting of retired batteries. However, existing technologies still have the following shortcomings: existing methods typically train each retired battery as an independent sample, using only the characteristics of a single battery to predict its health status, while ignoring the aging relationships within the retired battery group; sample relationships are static and cannot be continuously optimized as the model learns; and the graph topology lacks constraints on aging mechanisms.

[0005] Therefore, existing technologies primarily treat retired batteries as independent data samples, rather than considering the entire retired battery population as a continuously evolving aging relationship network. Thus, there is an urgent need to propose a novel method that can describe the aging relationships within a retired battery population, continuously optimize the graph topology based on deep learning results, and utilize dynamic graph structures to achieve intelligent sorting of retired lithium batteries. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a smart sorting method and system for retired lithium batteries based on the dynamic evolution of an aging relationship graph. This invention enables the relationship network between retired batteries to continuously evolve during model training, gradually approximating the actual aging relationship network in the graph structure, thus improving the ability to represent the health status of complex retired battery groups. After model training is complete, only a small number of labels are needed to complete the comprehensive utilization level classification, and further applications such as battery consistency grouping, health status ranking, and residual value assessment can be realized.

[0007] In a first aspect, this invention proposes a smart sorting method for retired lithium batteries based on dynamic evolution of aging relationship graphs, comprising: Step S1. Collect multi-source detection data characterizing the health status of retired lithium batteries to be sorted, perform linear interpolation and standardization on the multi-source detection data, and generate a node feature matrix including all retired lithium batteries to be sorted. Step S2. Calculate the aging similarity between each retired lithium battery node in all the retired lithium batteries to be sorted based on the multi-source detection data, and obtain the first aging correlation degree by establishing a weight vector, and construct the first aging relationship graph based on the first aging correlation degree; Step S3. Use the node feature matrix and the initialized aging relationship graph as input to the graph coding network to encode each retired lithium battery node, so as to output the deep aging features of all retired lithium battery nodes. Step S4. Based on the iterative update of the deep aging features and the first aging correlation, generate the updated second aging correlation, and construct the updated second aging relationship graph according to the second aging correlation. Step S5. If the second aging relationship graph satisfies the graph structure convergence criterion, then the converged third aging relationship graph is generated, and the intelligent sorting of the retired lithium batteries to be sorted is completed based on the third aging relationship graph.

[0008] As an optional implementation, the multi-source detection data includes the incremental capacity IC curve calculated from voltage and charge / discharge capacity, the differential voltage DV curve obtained from the unit charge / discharge capacity change, voltage response data, and AC internal resistance data.

[0009] As an optional implementation, step S2 specifically includes: All the retired lithium batteries to be sorted are formed into a retired lithium battery node set V. Each retired lithium battery in the node feature matrix corresponds to a node in the first aging relationship graph. , represented as , , This indicates the number of retired lithium batteries; Based on the multi-source detection data, the aging similarity among each node of the retired lithium battery in all the retired lithium batteries to be sorted was calculated. , including calculating the first Retired lithium battery nodes and the first IC curve similarity between retired lithium battery nodes DV curve similarity Voltage response similarity Similarity to internal resistance ; Based on the aging similarity and the established weight vector Obtain the first aging correlation. , represented as: ; For each node According to the first aging correlation Sort the nodes and retain the K nodes with the highest correlation as nodes. neighborhood And nodes Only with neighboring regions Nodes in Establish connections to construct the first aging relationship graph. ; in, , representing the connection relationship between nodes in the first aging relationship graph; , The edge weight represents the strength of the aging relationship between two retired lithium battery nodes.

[0010] As an optional implementation, the graph coding network can be any one of the following: graph attention network (GAT), graph convolutional network (GCN), or graph transformer network (Graph Transformer).

[0011] As an optional implementation, step S4 specifically includes: The semantic similarity between retired lithium battery nodes is obtained through iterative updates based on the deep aging features. Then, a linear scaling mapping is performed to obtain the mapped semantic similarity. : ; The first aging correlation is updated and the second aging correlation is generated according to the following formula. : ; For each node According to the second aging correlation Sort the nodes and retain the K nodes with the highest correlation as nodes. neighborhood According to the second aging correlation and neighboring areas Construct the updated second aging relationship graph ; t represents the iteration round number. The fusion coefficient is... , representing the connection relationship between nodes in the second aging relationship graph, edge weight , This represents the second aging correlation matrix. It is in the second aging correlation matrix A single element at a given position; if a node First entry into the neighboring area Then add a new connection. If the connection in the first aging graph does not enter the neighborhood... If the edge is not found, the corresponding edge is deleted, and the edge weights of the remaining edges are updated synchronously.

[0012] As an optional implementation, after generating the second aging relationship graph, the second aging relationship graph is input into a graph encoding network, and a self-supervised edge reconstruction loss is used as the training objective to optimize the node features in the node feature matrix, wherein the loss function is defined as: ; In the above formula, L is the loss function. Indicates the first Retired lithium battery nodes and the first Node pairs formed by retired lithium battery nodes The predicted probability of an aging association exists.

[0013] As an optional implementation, in the step of intelligent sorting of retired lithium batteries based on the third aging relationship diagram, intelligent sorting includes supervised classification intelligent sorting and unsupervised clustering intelligent sorting. In supervised classification intelligent sorting, a small number of retired lithium batteries that have completed feature detection are used as labeled samples and sorting results of different levels are output through a lightweight classifier. In unsupervised clustering intelligent sorting, retired lithium batteries with similar aging characteristics in the third aging relationship diagram are directly clustered and sorting results are output.

[0014] Secondly, this invention also proposes a smart sorting system for retired lithium batteries based on dynamic evolution of aging relationship diagrams, comprising: The node feature matrix generation module is used to collect multi-source detection data characterizing the health status of retired lithium batteries to be sorted, perform linear interpolation and standardization processing on the multi-source detection data, and generate a node feature matrix including all retired lithium batteries to be sorted. The first aging relationship graph generation module is used to calculate the aging similarity between each retired lithium battery node in all the retired lithium batteries to be sorted and retired based on the multi-source detection data, and to obtain the first aging correlation degree by establishing a weight vector, and to construct the first aging relationship graph based on the first aging correlation degree. The deep aging feature generation module is used to take the node feature matrix and the initial aging relationship graph as input to the graph coding network, and to encode each retired lithium battery node to output the deep aging features of all retired lithium battery nodes. The second aging relationship graph generation module is used to generate an updated second aging relationship degree based on the iterative update of the deep aging features and the first aging relationship degree, and to construct an updated second aging relationship graph based on the second aging relationship degree. The intelligent sorting output module is used to generate a converged third aging relationship graph if the second aging relationship graph satisfies the graph structure convergence criterion, and to complete the intelligent sorting of retired lithium batteries to be sorted based on the third aging relationship graph.

[0015] Thirdly, the present invention also proposes an electronic device, comprising: a processor and a memory; the memory storing a computer-readable program executable by the processor; the processor executing the computer-readable program to implement the steps in the intelligent sorting method for retired lithium batteries as described in the first aspect.

[0016] Fourthly, the present invention also proposes a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the intelligent sorting method for retired lithium batteries as described in the first aspect.

[0017] The beneficial effects of this invention are: (1) A dynamic evolution mechanism for aging relationship graphs is proposed for the first time. Instead of using a fixed topology, the graph structure can be continuously optimized based on the model learning results, thereby achieving adaptive updating of the relationship between retired battery groups.

[0018] (2) Establish a two-way collaborative optimization mechanism between graph structure and node representation. The graph structure determines the learning direction of node features, and the node features in turn correct the graph structure, avoiding the error accumulation caused by the fixed topology in traditional graph learning and improving the model's ability to express complex aging relationships.

[0019] (3) Make full use of retired battery group information. Compared with the traditional method of analyzing batteries one by one, this invention models retired batteries as a whole, which can fully explore the aging correlation between different batteries and improve the generalization ability across manufacturers, batches and service conditions.

[0020] (4) Reduce label dependence and detection costs. This invention utilizes graph structure learning to obtain high-quality node representations, requiring only a small number of labeled samples to complete the classification of retired batteries, reducing the number of capacity tests and improving engineering detection efficiency.

[0021] (5) Strong engineering applicability. This invention only relies on conventional data such as voltage, current, temperature, internal resistance and charge-discharge curves in the existing testing process. It does not require additional testing equipment and is easy to integrate into the intelligent testing platform and comprehensive utilization production line of retired batteries, and has good prospects for industrial application. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the intelligent sorting method for retired lithium batteries based on dynamic evolution of aging relationship diagram in an embodiment of the present invention; Figure 2 This is a schematic diagram of the dynamic evolution of the aging relationship graph in the intelligent sorting method for retired lithium batteries based on the dynamic evolution of the aging relationship graph in an embodiment of the present invention; Figure 3 This is a schematic diagram of the framework of the intelligent sorting system for retired lithium batteries based on dynamic evolution of aging relationship diagram in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0024] like Figures 1-2As shown, this embodiment provides a smart sorting method for retired lithium batteries based on dynamic evolution of aging relationship graphs, including: Step S1. Collect multi-source detection data characterizing the health status of retired lithium batteries to be sorted, perform linear interpolation and standardization on the multi-source detection data, and generate a node feature matrix including all retired lithium batteries to be sorted.

[0025] In this embodiment, when performing routine testing on the retired lithium batteries to be sorted and collecting relevant state data, the data includes charge / discharge voltage curves, charge / discharge current curves, surface temperature change curves, AC internal resistance, incremental capacity (IC) curves calculated from charge / discharge data (voltage and charge / discharge capacity), and differential voltage (DV) curves obtained from unit charge / discharge capacity changes. These data can respectively reflect the external operating state of the battery, polarization degree, and internal electrochemical aging characteristics, collectively describing the health status of the retired lithium batteries. It should be noted that, to achieve the purpose of this embodiment, the multi-source detection data in this embodiment includes four types of data: incremental capacity (IC) curves calculated from voltage and charge / discharge capacity, differential voltage (DV) curves obtained from unit charge / discharge capacity changes, voltage response data, and AC internal resistance data.

[0026] Because different detection devices have different sampling frequencies, each time series first uses linear interpolation to unify the sampling length, and then performs standardization processing: ; in: These are the original detection values; This is the mean of the feature; Standard deviation; This is a standardized result.

[0027] After processing, each retired battery is represented as a uniform node feature matrix: ; in: , The number of retired batteries; The duration; Represents the set of real numbers; The feature dimension is denoted as X. The node feature matrix X, which includes all retired lithium batteries to be sorted, is represented as: ; Use the node feature matrix X as input to the aging graph generated in the following steps.

[0028] Step S2. Calculate the aging similarity between each retired lithium battery node in all the retired lithium batteries to be sorted based on the multi-source detection data, and obtain the first aging correlation degree by establishing a weight vector, and construct the first aging relationship graph based on the first aging correlation degree.

[0029] In this embodiment, unlike traditional methods that treat all batteries as independent samples, this invention first establishes an aging relationship diagram describing the aging relationships of the entire retired battery group. The initial aging relationship diagram is defined as follows: ,in: Represents a set of nodes. This indicates the connection relationships between nodes. The edge weights are represented below. The following describes the specific construction process of the first aging relationship graph, focusing on the initial aging relationship graph.

[0030] All lithium batteries awaiting sorting and retirement are grouped into a retired lithium battery node set V. Each retired lithium battery in the node feature matrix X corresponds to a node in the first aging relationship graph. , represented as , , This indicates the number of retired lithium batteries; Based on the multi-source detection data, the aging similarity among each node of the retired lithium battery in all the retired lithium batteries to be sorted was calculated. , including calculating the first Retired lithium battery nodes and the first IC curve similarity between retired lithium battery nodes DV curve similarity Voltage response similarity Similarity to internal resistance , respectively represented as: ; ; ; ; in: , , , , , , , They represent the first IC curves, DV curves, voltage response sequences, and AC internal resistances of the block cell and the j-th cell. All similarities are normalized to [value missing]. The larger the value, the closer the aging state of the two batteries is.

[0031] In this embodiment, considering the varying importance of different aging features, learnable weights are used to fuse the four types of similarity. A weight vector is defined. Its value is determined by Parameterization determination: ; in: These are trainable parameters, initialized as zero vectors (i.e., the four features are equally weighted initially). During training, they are used together with the encoder parameters to generate a second aging relationship graph in step S4, which is then input into the encoder and backpropagated to optimize the loss function.

[0032] Based on the aging similarity and the established weight vector Obtain the first aging correlation. , represented as: ; in: , representing a node With nodes The initial aging correlation degree, i.e. the edge weight between two nodes. , representing the connection relationship between nodes in the first aging relationship graph; , The edge weight represents the strength of the aging relationship between two retired lithium battery nodes.

[0033] It should be noted that if all nodes in the old relationship graph are directly connected, the graph complexity will reach quadratic time complexity. The computational complexity is too high for tens of thousands of retired batteries in practical engineering. Therefore, this invention proposes a sparse construction strategy for the aging relationship graph: for each node... According to the first aging correlation Sort the nodes and retain the K nodes with the highest correlation as nodes. neighborhood And nodes Only with neighboring regions Nodes in Establish connections to ultimately form the first aging relationship diagram: ; Among them: edge weight .

[0034] In this embodiment, neighborhood Represented as: ; in: , Indicates retrieving the node The corresponding entire row adjacency vector, This means selecting the nodes corresponding to the K elements with the highest weights in a row. It should be noted that subsequent dynamic graph evolution will directly update this. Instead of recalculating the original features, this variable is retained throughout the training process. The generated first aging graph preserves local aging relationships while avoiding excessive graph density, thus improving the efficiency of subsequent training.

[0035] Step S3. Use the node feature matrix and the initial aging relationship graph as input to the graph coding network to encode each retired lithium battery node, so as to output the deep aging features of all retired lithium battery nodes.

[0036] In this embodiment, after establishing the first aging relationship graph, it is necessary to learn the hidden aging features of each node. This embodiment employs a graph coding network. Encode the nodes, with node features as input. Relationship with first aging diagram The output is a low-dimensional representation of the nodes: ; in, This represents the overall output of the graph coding network Encoder. This represents the set of deep aging characteristics of N retired batteries. Each Indicates the first The deep aging characteristics of retired batteries are analyzed. The encoder here is not limited to a specific network structure; it can employ Graph Attention Network (GAT), Graph Convolutional Network (GCN), GraphTransformer, or other graph representation learning models. The purpose of this embodiment is not to encode the network itself, but to dynamically update the aging graph based on node representations. Therefore, any graph encoder capable of learning node representations can be used as an implementation. At this point, the model has completed the first mapping from the original multi-source detection data to the first aging graph, and then to the deep node representation, providing input for the subsequent dynamic evolution of the aging graph.

[0037] Step S4. Based on the iterative update of the deep aging features and the first aging correlation, generate the updated second aging correlation, and construct the updated second aging relationship graph according to the second aging correlation.

[0038] In this embodiment, existing graph learning methods typically construct a fixed graph structure before training begins and maintain the graph topology unchanged throughout the training process. However, the initial graph structure is constructed based only on shallow physical features, making it difficult to accurately reflect the true aging relationships of retired batteries. When erroneous connections exist in the graph, these erroneous relationships will continue to propagate throughout the training process, affecting the learning effect of node features. To address the above problems, this application proposes a dynamic evolution algorithm for aging relationship graphs. The core idea is to re-evaluate the relationships between nodes using the node depth representations obtained in the current training round and dynamically update the graph topology, so that the graph structure continuously approximates the real aging relationship network as the model trains.

[0039] (1) Node embedding similarity calculation. The deep aging features of each node obtained from step S3 are represented as follows: ,in This represents the current iteration round number. The semantic similarity between nodes is calculated based on their node embeddings. ; in: This indicates the degree of node similarity learned by the model. This represents the deep aging feature of the i-th node after t iterations. This represents the deep aging feature of the j-th node after t iterations. Since this similarity is derived from the representation of deep features, it can reflect the potential aging patterns that traditional physical features cannot express. To maintain consistency with the aging correlation range [0,1], it is linearly scaled and mapped to... , represented as: ; (2) Aging correlation update. Initial aging correlation Derived from physical detection data, as training progresses, the node relationships obtained from deep learning should be gradually integrated. Therefore, the second aging correlation is obtained by updating the first aging correlation using the following formula. : ; in, This represents the aging correlation degree of the first aging correlation at iteration t. This is the fusion coefficient. In the early stages of training, the node representations are not yet stable, so we take... The correlation mainly depends on physical features; as node representations mature in the later stages of training, the correlation gradually decreases. This allows the graph structure to be updated based more on deep aging characteristics, resulting in the updated aging correlation matrix, i.e., the second aging correlation matrix. This serves as the basis for the next round of graph topology updates. To avoid the quadratic time complexity caused by full computation... The cost and correlation update are calculated only for candidate node pairs (the union of the current neighborhoods of each node and randomly sampled candidate nodes).

[0040] (3) Graph topology reconstruction. This embodiment not only updates edge weights, but also reconstructs the graph topology based on the new aging correlation, including three operations for each node: adding edges, deleting edges, and adjusting weights. According to the second aging correlation Sort the nodes and retain the K nodes with the highest correlation as nodes. neighborhood , ; According to the second aging correlation and neighboring areas Construct the updated second aging relationship graph Where t represents the number of iterations, The fusion coefficient is... , representing the connection relationship between nodes in the second aging relationship graph, edge weight , This represents the second aging correlation matrix. It is in the second aging correlation matrix A single element at a given position; if a node First entry into the neighboring area Then add a new connection. If the connection in the first aging graph does not enter the neighborhood... If the edge is not found, the corresponding edge is deleted, and the edge weights of the remaining edges are updated synchronously. Unlike existing fixed graph learning methods, the graph topology in this invention can continuously evolve as the model trains, making the graph structure increasingly approximate the aging relationship of real retired batteries.

[0041] Step S5. If the second aging relationship graph satisfies the graph structure convergence criterion, then the converged third aging relationship graph is generated, and the intelligent sorting of the retired lithium batteries to be sorted is completed based on the third aging relationship graph.

[0042] In this embodiment, after generating the second aging relationship graph, the sorting process is not initiated directly. Instead, the new aging relationship graph is reused for graph representation learning. The second aging relationship graph is input into the graph encoding network, and a self-supervised edge reconstruction loss is used as the training objective to optimize the node features in the node feature matrix. The loss function is defined as follows: ; In the above formula, Indicates the first Retired lithium battery nodes and the first Node pairs formed by retired lithium battery nodes The predicted probability of an aging association is expressed as: ; in: For the Sigmoid function; For temperature coefficient, each Indicates the first The deep aging characteristics of each retired battery Indicates the first The deep aging characteristics of a retired battery block, where d represents the dimension. Let represent a d-dimensional real vector space.

[0043] The summation iterates through the current second aging graph, traversing both connected and sampled non-connected node pairs. This loss-driven node represents an accurate reconstruction of the aging graph, and simultaneously optimizes the encoder parameters and fusion weights through backpropagation. Physical features that are more consistent with the learned aging structure will receive greater weight. Thus, graph structure, node representation, and feature weights are optimized synergistically under the same training objective.

[0044] To determine whether a graph is stable, a graph structure convergence criterion is introduced. This involves defining the change in the incidence matrix over two consecutive rounds and determining whether the convergence condition is met: ; in, This represents the change in the correlation matrix at round t. This represents the aging correlation matrix at iteration t. This represents the updated aging correlation matrix: Denotes the Frobenius norm; A preset convergence threshold is set. If the above formula is satisfied, the aging graph is considered to be basically stable; otherwise, steps S3 and S4 are continued. To prevent oscillations caused by discrete edge addition and deletion operations, a maximum number of iteration rounds is also set. The iteration is forcibly stopped after reaching this number of rounds. Finally, the converged third aging relationship graph is obtained. , Denotes the set of edges after convergence. This represents the set of edge weights after convergence. At this point, the graph structure can more accurately describe the aging relationship network of the entire retired battery group.

[0045] After the graph structure converges, the graph encoding network parameters are frozen, and the final nodes are used for representation. As a characteristic of the health status of retired batteries, intelligent sorting can be achieved in two ways depending on the actual application requirements.

[0046] The first approach is supervised classification. A small number of retired batteries that have undergone capacity testing are selected as labeled samples, and a lightweight classifier is trained using node representations as input. ; The output categories include comprehensive utilization levels such as Grade A, Grade B, and Grade C. Batteries with poor health conditions that do not meet the requirements for continued use are placed into the dismantling and recycling (reuse) path. Since the node representation has fully incorporated the group aging relationship, only a small number of tags are needed to complete the classification, significantly reducing the cost of capacity testing.

[0047] The second method is unsupervised clustering. (Directly...) Cluster analysis is performed to obtain multiple aging clusters, and then combined with a small amount of manual calibration, the comprehensive utilization level classification can be completed.

[0048] Furthermore, based on the edge weights in the final aging relationship graph It can also directly calculate the consistency between batteries, providing a basis for grouping in comprehensive utilization scenarios: the higher the edge weight, the closer the aging paths of the two batteries are, and the more suitable they are to be grouped into the same battery module; the lower the edge weight, the less suitable they are to be grouped, so as to reduce the consistency mismatch problem in the subsequent operation process.

[0049] Based on the above-mentioned intelligent sorting method for retired lithium batteries based on dynamic evolution of aging relationship graphs, this embodiment of the invention also provides an intelligent sorting system 30 for retired lithium batteries based on dynamic evolution of aging relationship graphs, such as... Figure 3 As shown, it includes: The node feature matrix generation module 301 is used to collect multi-source detection data characterizing the health status of retired lithium batteries to be sorted, perform linear interpolation and standardization processing on the multi-source detection data, and generate a node feature matrix including all retired lithium batteries to be sorted. The first aging relationship graph generation module 302 is used to calculate the aging similarity between each retired lithium battery node in all retired lithium batteries to be sorted based on the multi-source detection data, and obtain the first aging correlation degree by establishing a weight vector, and construct the first aging relationship graph based on the first aging correlation degree. The deep aging feature generation module 303 is used to take the node feature matrix and the initial aging relationship graph as input to the graph coding network, and to encode each retired lithium battery node to output the deep aging features of all retired lithium battery nodes. The second aging relationship graph generation module 304 is used to generate an updated second aging relationship degree based on the iterative update of the deep aging features and the first aging relationship degree, and to construct an updated second aging relationship graph based on the second aging relationship degree. The intelligent sorting output module 305 is used to generate a converged third aging relationship graph if the second aging relationship graph satisfies the graph structure convergence criterion, and to complete the intelligent sorting of retired lithium batteries to be sorted based on the third aging relationship graph.

[0050] Based on the above-described intelligent sorting method for retired lithium batteries using dynamic evolution of the aging relationship diagram, this embodiment also provides an electronic device, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing electronic device. The electronic device includes a processor, memory, and display.

[0051] In some embodiments, the memory can be an internal storage unit of the electronic device, such as a hard drive or RAM. In other embodiments, the memory can be an external storage device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory can include both internal and external storage devices. The memory is used to store application software and various types of data installed on the electronic device, such as program code installed on the electronic device. The memory can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory stores a retired lithium battery intelligent sorting program, which can be executed by a processor to implement the retired lithium battery intelligent sorting method of the embodiments of this application.

[0052] In some embodiments, the processor may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in memory or process data, such as executing a smart sorting method for retired lithium batteries.

[0053] In some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display is used to show information from the intelligent sorting electronics for retired lithium batteries and to display a visual user interface. Components of the electronic device communicate with each other via a system bus.

[0054] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0055] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent sorting method for retired lithium batteries based on dynamic evolution of aging relationship graph, characterized in that, The method comprises: Step S1. Collecting multi-source detection data representing the health state of the retired lithium batteries to be sorted, performing linear interpolation and standardization processing on the multi-source detection data, and generating a node feature matrix comprising all retired lithium batteries to be sorted; Step S2. Calculating the aging similarity between each retired lithium battery node in all retired lithium batteries to be sorted based on the multi-source detection data, and obtaining a first aging correlation degree by establishing a weight vector, and constructing a first aging relationship graph according to the first aging correlation degree; Step S3. Taking the node feature matrix and the initialized aging relationship graph as inputs of a graph encoding network, and using the graph encoding network to encode each retired lithium battery node to output deep aging features of all retired lithium battery nodes; Step S4. Based on the iterative update of the deep aging features and the first aging correlation degree, generating an updated second aging correlation degree, and constructing an updated second aging relationship graph according to the second aging correlation degree; Step S5. If the second aging relationship graph meets the graph structure convergence criterion, a third aging relationship graph is generated, and the intelligent sorting of the retired lithium batteries to be sorted is completed based on the third aging relationship graph.

2. The method of claim 1, wherein the method is characterized by, The multi-source detection data comprises an incremental capacity IC curve calculated from voltage and charge-discharge capacity, a differential voltage DV curve obtained from unit charge-discharge capacity change, voltage response data, and alternating current resistance data.

3. The method of claim 2, wherein the method further comprises: Step S2 specifically comprises: All the retired lithium batteries to be sorted form a retired lithium battery node set V, each piece of retired lithium battery in the node feature matrix corresponds to a node in the first aging relationship graph , is expressed as , , represents the number of retired lithium batteries; calculate aging similarity between each retired lithium battery node in all retired lithium batteries to be sorted based on the multi-source detection data , including calculating the IC curve similarity between the retired lithium battery node and the first retired lithium battery node , the DV curve similarity , the voltage response similarity , and the internal resistance similarity between the retired lithium battery node and the first retired lithium battery node ; According to the aging similarity and the established weight vector Obtaining the first aging correlation degree is expressed as: ; For each node , according to the first aging correlation , the top K nodes with the largest correlation are reserved as the neighborhood of the node , and the node only establishes a connection with the nodes in the neighborhood , thereby constructing a first aging relationship graph ; wherein, , represents the connection relationship between nodes in the first aging relationship graph; , represents the edge weight, used to represent the strength of the aging relationship between two retired lithium battery nodes.

4. The method of claim 1, wherein the method is characterized by, The graph encoding network is any one of a graph attention network GAT, a graph convolution network GCN, or a graph transformer network GraphTransformer.

5. The method of claim 3, wherein the method further comprises: Step S4 specifically comprises: The semantic similarity between the retired lithium battery nodes is obtained based on iterative updating of the deep aging features And linear scaling mapping is performed to obtain the mapped semantic similarity : ; The first aging correlation is updated to generate a second aging correlation according to the following equation : ; For each node , according to the second aging correlation , the first K nodes with the largest correlation are reserved as the neighborhood of the node , and the second aging relationship graph is constructed according to the second aging correlation and the neighborhood . ​ t represents the number of iteration rounds, is the fusion coefficient, represents the connection relationship between nodes in the second aging relationship graph, and the edge weight , represents the second aging correlation degree matrix, is a single element in the second aging correlation degree matrix at a position; if a node first enters the neighborhood , a new connection is added ; if the connection in the first aging relationship graph does not enter the neighborhood , the corresponding edge is deleted, and the remaining edge is updated synchronously.

6. The method of claim 5, wherein the method further comprises: After generating the second aging relationship graph, the second aging relationship graph is input into the graph encoding network, and a self-supervised edge reconstruction loss is used as a training target to optimize the node features in the node feature matrix, wherein the loss function is defined as: ; In the above formula, L is a loss function, representing the retired lithium battery nodes and the retired lithium battery nodes constitute a node pair There is a prediction probability associated with aging.

7. The method of claim 5, wherein the method further comprises: In the step of completing the intelligent sorting of the retired lithium batteries to be sorted based on the third aging relationship graph, the intelligent sorting comprises supervised classification intelligent sorting and unsupervised clustering intelligent sorting, wherein in the supervised classification intelligent sorting, a small number of retired lithium batteries with completed feature detection are used as labeled samples to output sorting results of different grades through a lightweight classifier; and in the unsupervised clustering intelligent sorting, retired lithium batteries with similar aging characteristics in the third aging relationship graph are directly clustered to output the sorting results.

8. An intelligent retired lithium battery sorting system based on the dynamic evolution of the aging relationship graph, characterized in that, Comprise: A node feature matrix generation module for collecting multi-source detection data representing the health state of the retired lithium batteries to be sorted, performing linear interpolation and standardization processing on the multi-source detection data, and generating a node feature matrix comprising all retired lithium batteries to be sorted; A first aging relationship graph generation module for calculating the aging similarity between each retired lithium battery node in all retired lithium batteries to be sorted based on the multi-source detection data, and obtaining a first aging correlation degree by establishing a weight vector, and constructing a first aging relationship graph according to the first aging correlation degree; A graph encoding network for encoding each retired lithium battery node based on the node feature matrix and the initialized aging relationship graph to output deep aging features of all retired lithium battery nodes; a deep aging feature generation module configured to input the node feature matrix and the initialized aging relationship graph into a graph encoding network to encode each retired lithium battery node and output deep aging features of all the retired lithium battery nodes; a second aging relationship graph generation module configured to generate updated second aging correlation degrees based on iterative updates of the deep aging features and the first aging correlation degrees, and construct an updated second aging relationship graph according to the second aging correlation degrees; an intelligent sorting output module configured to generate a converged third aging relationship graph if the second aging relationship graph satisfies a graph structure convergence criterion, and complete intelligent sorting of the retired lithium batteries to be sorted based on the third aging relationship graph.

9. An electronic device, comprising: comprising: a processor and a memory; the memory stores a computer readable program executable by the processor; the processor implements the steps in the intelligent sorting method of the retired lithium batteries according to any one of claims 1-7 when executing the computer readable program.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores one or more programs executable by one or more processors to implement the steps in the intelligent sorting method of the retired lithium batteries according to any one of claims 1-7.