Retired power battery echelon configuration optimization method based on full life cycle of battery cell

Abstract

The application discloses a retired power battery gradient configuration optimization method based on a whole life cycle of an electric core, and belongs to the technical field of power battery utilization. Historical operation images of the whole service cycle of the retired power battery are collected, a labeled historical image set is generated by performing health degree clustering labeling, training historical image set is obtained by performing key sampling time decision preprocessing on the set, an electric core gradient configuration clustering network is inputted to construct an aging gradient clustering configuration model, a gradient clustering block is performed on current images of a battery to be configured to obtain degradation state detection results of each block, an overall health risk index is calculated by fusing and integrating each block detection result, and an optimal configuration report is generated by selecting the lowest risk electric core area in combination with a matching relationship of an expected service cycle. The application realizes adaptive optimization of the gradient configuration scheme, significantly improves the robustness and reliability of the configuration decision under long-term service conditions, and effectively reduces the early failure risk and safety hazards of the recombined battery module.

Description

Technical Field

[0001] This invention discloses a method for optimizing the tiered configuration of retired power batteries based on the entire life cycle of battery cells, belonging to the field of power battery utilization technology. Background Technology

[0002] With the large-scale promotion and application of electric vehicles and energy storage systems, power batteries still have a high remaining capacity after reaching their on-board service life. The cascade utilization of retired power batteries has become an important way to recycle resources and reduce energy storage costs. In the process of reconfiguring retired power batteries, because different cells in the same battery pack have experienced different operating temperatures, charge / discharge rates, and states of charge ranges during long-term service, the aging degree and degradation trajectory of each cell show significant inconsistencies, resulting in the overall performance of the battery pack being severely constrained by the weakest cells.

[0003] Existing tiered configuration methods typically rely on offline capacity testing and internal resistance measurement of retired batteries. Based on the test results at a single point in time, the cells are simply graded, screened, and regrouped. This static evaluation method ignores the temporal patterns and dynamic characteristics of cell aging evolution throughout its entire life cycle, making it difficult to accurately depict the degradation trend of each cell in subsequent tiered services. This results in the configuration scheme experiencing premature performance degradation or even safety failure in actual operation.

[0004] In the fields of computer vision and image processing, some studies have attempted to use operational monitoring images of power batteries for the detection of physical defects such as cell deformation, leakage marks, and tab corrosion. However, these methods often treat images as independent detection samples, extracting only the spatial visual features of the current frame, without systematically modeling and analyzing the temporal image sequence of the cell throughout its entire life cycle. Furthermore, most existing image clustering and gradient partitioning methods are based on fixed feature extraction templates and preset threshold rules, lacking the ability to adaptively adjust to feature distribution drift during cell aging. When the operating conditions of the cell or the image acquisition environment change, the stability and generalization of the clustering results decrease significantly. At the same time, current optimization decisions for tiered configurations are usually based on individual cells for independent evaluation and selection, failing to consider the impact of topological coupling relationships and spatial position weights between cells on overall performance from the perspective of the overall battery pack structure. This results in significant deficiencies in the regional synergy and engineering feasibility of the configuration results. Summary of the Invention

[0005] To address the aforementioned shortcomings, this invention proposes a method for optimizing the tiered configuration of retired power batteries based on the entire lifecycle of battery cells. This method involves collecting historical operational images of retired power batteries throughout their service life and performing health-related clustering annotations to construct a labeled image set covering the entire time-series information of cell aging. Based on this, a key sampling moment decision preprocessing strategy is designed to filter images from dense time-series images at turning points where aging characteristics significantly change, effectively reducing the data scale for subsequent processing while retaining key degradation information. Furthermore, a tiered configuration clustering network for battery cells is constructed and iteratively trained, enabling the aging gradient clustering configuration model to adaptively track the distribution changes of cell characteristics throughout the lifecycle, avoiding classification bias caused by fixed templates. In the application phase, this invention uses this model to perform gradient clustering on the current image of the battery to be configured, obtaining the degradation state detection results for each cell block. Through fusion and integration processing, the block-level detection results are combined with the physical location weights and electrical connection topology weights of the cells to calculate the overall health risk index. Finally, based on the matching relationship between the health risk index and the expected service life, the cell region with the lowest risk is selected to output a configuration report. The above-mentioned scheme extracts dynamic aging patterns from time-series images throughout the entire lifecycle, drives the adaptive updating of the clustering model through iterative iteration, and performs structured fusion of block detection results. This effectively overcomes the limitations of static single-point detection and independent evaluation, and significantly improves the robustness and reliability of tiered configuration decisions under long-term service conditions.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] A method for optimizing the tiered configuration of retired power batteries based on the entire life cycle of battery cells includes:

[0008] S1. Collect historical operating images corresponding to multiple retired power batteries, perform health clustering and labeling processing on each historical operating image, and generate a set of labeled historical images of battery cells.

[0009] S2, perform key sampling moment decision preprocessing on the cell-annotated historical image set, complete image specification unification processing and signal noise suppression processing, and obtain cell training historical image set;

[0010] S3, feed the set of battery cell training history images into the battery cell tiered configuration clustering network to perform iterative processing and construct a battery cell aging gradient clustering configuration model.

[0011] S4. Using the cell aging gradient clustering configuration model, perform gradient clustering block processing on the current operating image of the power battery to be configured for retirement, obtain the degradation state detection results of each cell block image to be configured, perform fusion and integration processing on the degradation state detection results of each cell block image to be configured, and calculate the health risk index of the power battery to be configured for retirement.

[0012] S5. Based on the matching relationship between the health risk index and the expected service life, select the cell area with the lowest health risk index from each cell block area of ​​the power battery to be configured for retirement as the output, and generate a risk optimization configuration report.

[0013] Furthermore, S1 collects historical operational images of each of the multiple retired power batteries, specifically:

[0014] Surface image data and internal structure image data of each retired power battery at multiple time points during its historical service phase are acquired. The surface image data and internal structure image data of the same retired power battery at the same time point are spatiotemporally aligned to form aligned image pairs of each retired power battery at multiple time points. The historical operation image sequence of each retired power battery is constructed based on the aligned image pairs.

[0015] Furthermore, S1 performs health clustering annotation processing on each of the historical operating images to generate a set of cell-annotated historical images, specifically:

[0016] Extract the cell outline boundaries from each of the historical operation images, and cut each historical operation image into several cell sub-images based on the cell outline boundaries.

[0017] Image feature vectors characterizing the aging degree of each cell sub-image are extracted. The image feature vectors are input into the health clustering space for clustering operations. Based on the clustering operation results, the health level label corresponding to each cell sub-image is determined. The cell sub-images carrying the health level label are then aggregated into a set of labeled historical images of the cells.

[0018] Furthermore, S2 specifically refers to:

[0019] Obtain the original acquisition timestamps corresponding to each image in the battery cell labeled historical image set, and determine the key sampling time sequence based on the distribution density of each original acquisition timestamp on the time axis;

[0020] Using the key sampling time sequence as an index, the labeled images that match each key sampling time are extracted from the set of labeled historical images of the battery cell to form a subset of key sampling images;

[0021] Spatial resolution alignment processing is performed on each image in the key sampled image subset to give each image a uniform spatial pixel scale;

[0022] Temporal noise filtering is performed on each image after spatial resolution alignment to suppress random temporal fluctuation components introduced during image acquisition.

[0023] The images, after spatial resolution alignment and temporal noise filtering, are compiled into a set of historical images for battery cell training.

[0024] Furthermore, S3 specifically refers to:

[0025] S31, the set of battery cell training history images is divided into several training batches, and each training batch contains an equal number of images.

[0026] S32, input the images of the current training batch into the cell tiered configuration clustering network, the cell tiered configuration clustering network extracts aging feature representations for each image, and assigns each image to the corresponding aging gradient clustering cluster according to the similarity measure between the aging feature representations;

[0027] S33, After completing the allocation of the current training batch, update the cluster center features of each cluster in the cell tiered configuration clustering network according to the allocation results;

[0028] S34, use the updated cluster center features as the initial cluster centers for the next training batch, repeat the image allocation and cluster center update operations until all training batches are processed, and output the cell aging gradient clustering configuration model.

[0029] Furthermore, S32 specifically includes:

[0030] The cell tiered configuration clustering network performs multi-layer feature abstraction operations on each image, extracting multi-scale aging feature representations from each image, ranging from local texture features to global degradation distribution features;

[0031] Calculate the distance metric between the multi-scale aging feature representation of each image and the cluster center features of each current cluster;

[0032] Based on the distance metric, the target cluster with the smallest distance metric to each image is determined, and each image is assigned to the corresponding target cluster.

[0033] Furthermore, in step S4, the cell aging gradient clustering configuration model is used to perform gradient clustering block processing on the current operating image of the retired power battery to be configured, and the degradation state detection results of each cell block image to be configured are obtained, specifically:

[0034] Obtain the current operating image of the retired power battery to be configured at the current moment, and cut the current operating image into several cell block images to be configured according to the physical arrangement structure of the cells.

[0035] Each of the cell block images to be configured is input into the cell aging gradient clustering configuration model one by one, and the cell aging gradient clustering configuration model determines the gradient clustering position of each cell block image to be configured in the aging gradient space.

[0036] Based on the gradient clustering position of each cell block image to be configured, the degradation state description parameters corresponding to each gradient clustering position are retrieved from the cell aging gradient clustering configuration model, and the degradation state description parameters are output as the degradation state detection results of each cell block image to be configured.

[0037] Furthermore, in S4, the degradation state detection results of each of the cell blocks to be configured are fused and integrated to calculate the health risk index of the retired power battery to be configured, specifically:

[0038] Obtain the degradation state detection results of each cell block image to be configured, wherein the degradation state detection results include the state description values ​​of each cell block in multiple degradation dimensions;

[0039] Based on the physical location weight and electrical connection topology weight of each cell block in the decommissioned power battery to be configured, the state description values ​​of each cell block in multiple degradation dimensions are weighted and aggregated to obtain the overall state aggregate value of the decommissioned power battery to be configured in each degradation dimension.

[0040] The overall state aggregation value on each degradation dimension is mapped to the health risk assessment space, and the health risk index of the power battery to be configured for retirement is calculated through the risk contribution weight corresponding to each degradation dimension in the health risk assessment space.

[0041] Furthermore, S5 specifically includes:

[0042] Obtain the cycle length parameter and working condition intensity parameter corresponding to the expected service cycle, and determine the health risk index threshold based on the cycle length parameter and working condition intensity parameter;

[0043] The health risk index of each cell block region in the power battery to be configured is compared with the health risk index threshold, and candidate cell block regions with health risk indices lower than the health risk index threshold are selected.

[0044] Select the cell block region with the lowest health risk index from the candidate cell block regions as the target configuration region, and generate a risk optimization configuration report that includes the spatial location information of the target configuration region, the health risk index value, and the matching degree evaluation information with the expected service life.

[0045] Furthermore, after generating the risk-optimized configuration report, it also includes:

[0046] The risk optimization configuration report is output to the tiered configuration decision terminal. The tiered configuration decision terminal generates a cell reorganization operation instruction for the retired power battery to be configured based on the spatial location information of the target configuration area in the risk optimization configuration report. The cell reorganization operation instruction includes the cell area identifier to be removed and the cell area identifier to be retained.

[0047] This invention discloses a method for optimizing the tiered configuration of retired power batteries based on the entire life cycle of battery cells, belonging to the field of power battery utilization technology. The method involves collecting historical operational images of retired power batteries throughout their entire service life, performing health-level clustering annotation to generate an annotated historical image set, performing key sampling moment decision preprocessing on the set to obtain a training historical image set, inputting it into a cell tiered configuration clustering network to construct an aging gradient clustering configuration model, performing gradient clustering block segmentation on the current image of the battery to be configured to obtain the degradation state detection results of each block, fusing and integrating the detection results of each block to calculate the overall health risk index, and combining the matching relationship with the expected service life to select the cell region with the lowest risk to generate an optimized configuration report. This invention achieves adaptive optimization of the tiered configuration scheme, significantly improving the robustness and reliability of configuration decisions under long-term service conditions, and effectively reducing the early failure risk and safety hazards of reassembled battery modules. Attached Figure Description

[0048] Figure 1 A flowchart illustrating the workflow of a method for optimizing the tiered configuration of retired power batteries based on the entire life cycle of battery cells, as claimed in this embodiment of the invention.

[0049] Figure 2 The second flowchart is a method for optimizing the tiered configuration of retired power batteries based on the entire life cycle of battery cells, as claimed in the embodiments of the present invention.

[0050] Figure 3 The third flowchart is a method for optimizing the tiered configuration of retired power batteries based on the entire life cycle of the battery cell, as claimed in the embodiments of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0052] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications in the embodiments of this application, such as up, down, left, right, front, back, etc., are only used to explain the relative positional relationships and movements between components in a specific orientation as shown in the accompanying drawings. If the specific orientation changes, the directional indications will change accordingly. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0053] References to embodiments herein mean that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0054] According to the first embodiment of the present invention, referring to Figure 1 This invention claims protection for a method for optimizing the tiered configuration of retired power batteries based on the entire life cycle of battery cells, comprising:

[0055] S1. Collect historical operating images corresponding to multiple retired power batteries, perform health clustering and labeling processing on each historical operating image, and generate a set of labeled historical images of battery cells.

[0056] S2, perform key sampling moment decision preprocessing on the cell-annotated historical image set, complete image specification unification processing and signal noise suppression processing, and obtain cell training historical image set;

[0057] S3, feed the set of battery cell training history images into the battery cell tiered configuration clustering network to perform iterative processing and construct a battery cell aging gradient clustering configuration model.

[0058] S4. Using the cell aging gradient clustering configuration model, perform gradient clustering block processing on the current operating image of the power battery to be configured for retirement, obtain the degradation state detection results of each cell block image to be configured, perform fusion and integration processing on the degradation state detection results of each cell block image to be configured, and calculate the health risk index of the power battery to be configured for retirement.

[0059] S5. Based on the matching relationship between the health risk index and the expected service life, select the cell area with the lowest health risk index from each cell block area of ​​the power battery to be configured for retirement as the output, and generate a risk optimization configuration report.

[0060] In this embodiment, when collecting historical operating images of multiple retired power batteries, for each retired power battery, operating monitoring images during the charge-discharge cycle are retrieved from the image storage server associated with the battery management system at fixed time intervals throughout the entire time span from the date of its initial installation and use to the date of its final decommissioning. For each retired power battery, the collected historical operating images cover its entire service life cycle, from the initial installation and power-on to the final decommissioning and power-off. The time interval is set to collect one frame of image after each complete charge-discharge cycle. If the battery is in a static or idle state during a certain period, a frame of static state image is collected every 24 hours. After completing the image collection of all retired power batteries, health clustering labeling processing is performed on each frame of historical operating image. This processing first... First, the individual battery cells contained in each image frame are identified. Then, based on the physical morphology and color distribution characteristics of the image area where each battery cell is located, image frames of the same battery cell at different time points are associated with the same battery cell identifier. Next, using the battery cell identifier as the unit, all image frames of that battery cell from the first acquisition to the last acquisition are arranged in chronological order, and a health label is assigned to each image frame. This label is determined based on the cycle number interval corresponding to the image frame and the measured capacity retention rate data of the battery cell at that cycle number. The image frames of each battery cell carrying the health label are classified and stored according to the battery cell identifier, forming a set of labeled historical images of battery cells with individual battery cells as the index unit and time frames as the sorting basis. Each image frame in this set is accompanied by metadata information in four dimensions: unique battery cell code, acquisition timestamp, cycle number count, and health label.

[0061] When performing key sampling moment decision preprocessing on the cell-annotated historical image set, the loop count metadata attached to each frame of the set is first traversed to count the loop count increment of each cell between adjacent acquisition frames. When the loop count increment of a frame exceeds a preset increment threshold, the time point corresponding to that frame is marked as a candidate key sampling moment. Simultaneously, when the health label of a frame changes level compared to the previous frame, the time point corresponding to that frame is also marked as a candidate key sampling moment. After summarizing all candidate key sampling moments, they are arranged in chronological order, and redundant candidate moments with time intervals less than the lower limit of the acquisition interval are deleted. The remaining candidate key sampling moments are retained as the final key sampling moment sequence. The key sampling time sequence is formed by extracting image frames corresponding to each key sampling time from the cell-annotated historical image set. For each image in the key sampling image subset, image standardization processing is performed. Signal and noise suppression processing is then performed on each image after standardization. This processing uses spatial filtering to locally smooth the pixel channel values ​​in each frame to eliminate random noise components introduced by illumination fluctuations, sensor thermal noise, and transmission interference during image acquisition. After noise suppression processing, the pixel channel values ​​of each image remain unchanged within the integer range of 0 to 255. All image frames after standardization and signal and noise suppression processing are reorganized according to their original acquisition time order to form the cell training historical image set.

[0062] When feeding the battery cell training history image set into the battery cell tiered configuration clustering network for iterative processing, the initial number of cluster centers in the network is set to the preset total number of gradient categories. The feature vector of each initial cluster center is initialized by randomly selecting the feature descriptors of a corresponding number of image frames from the battery cell training history image set. After receiving each image frame, the network first divides the image frame into spatial regions, extracts the texture response intensity and edge direction distribution in each region as the original features of the image frame, and then maps the original features to the aging gradient feature space through multiple feature transformation operations. In the aging gradient feature space, the distance between the image frame and each cluster center is calculated, and the image frame is divided into the corresponding aging gradient according to the minimum distance principle. Gradient clustering: After each frame of image is divided, the cluster center feature vector of the cluster to which the frame belongs is recalculated, that is, the component mean of the feature vectors of all image frames in the cluster is used as the updated cluster center. After all image frames in the battery cell training history image set have been divided and the cluster center updated, the change in cluster center after this complete traversal is recorded. If the change in cluster center is greater than the preset convergence threshold, the input order of all image frames is shuffled again and the next round of traversal iteration is started. The iteration stops when the change in cluster center in two consecutive rounds is less than or equal to the convergence threshold. After the iteration stops, the final feature vector of each cluster center and the list of image frame numbers contained in each cluster are fixed and stored as the battery cell aging gradient clustering configuration model.

[0063] When performing gradient clustering and block processing on the current operating image of the decommissioned power battery to be configured using the cell aging gradient clustering configuration model, the real-time operating image of the decommissioned power battery to be configured is first acquired. The acquisition conditions of this real-time operating image are consistent with those of the historical operating images. The real-time operating image is then divided into several rectangular blocks according to the actual number of rows and columns of the cells in the decommissioned power battery to be configured. Each block corresponds to the imaging area of ​​a single cell, and the spatial dimensions of each block are consistent, occupying a fixed pixel width horizontally and a fixed pixel height vertically. Each block image is input into the cell aging gradient clustering configuration model. The model calculates the distance between the feature vector of the block image and each cluster center, and assigns the block image to the aging gradient category corresponding to the nearest cluster center. The degradation state description parameters associated with the aging gradient category are used as the degradation state detection result of the block image. These degradation state description parameters include the lower limit of the capacity decay range corresponding to the category. The upper and lower limits of the internal resistance growth range and the lower and upper limits of the self-discharge rate range are calculated. When performing fusion and integration processing on the degradation state detection results of each cell block image to be configured, different weighting coefficients are assigned according to the series and parallel positions of each block in the battery physical structure. The weighting coefficient of the block closer to the battery output end in the series position is higher, and the weighting coefficient of the block in the middle layer in the parallel position is higher than that of the block in the edge layer. The capacity decay range, internal resistance growth range and self-discharge rate range of each block are multiplied by the corresponding weighting coefficients and summed, and then divided by the sum of the weighting coefficients to obtain the overall capacity decay aggregate value, overall internal resistance growth aggregate value and overall self-discharge rate aggregate value of the retired power battery to be configured. The overall capacity decay aggregate value, overall internal resistance growth aggregate value and overall self-discharge rate aggregate value are converted into capacity risk component, internal resistance risk component and self-discharge risk component respectively according to their respective health risk mapping curves. The three risk components are added together to obtain the health risk index.

[0064] When selecting the cell region with the lowest risk index based on the matching relationship between the health risk index and the expected service life, the expected number of cycles corresponding to the user-input expected service life is obtained. Based on this expected number of cycles, the maximum allowable health risk index is found from a preset cycle-risk comparison table. The health risk index of each cell block region of the battery to be configured is calculated one by one. The calculation method is to extract the capacity decay interval, internal resistance increase interval, and self-discharge rate interval values ​​from the degradation state detection results of the cell block region, map them to the capacity risk component, internal resistance risk component, and self-discharge rate component respectively, and then sum them. The health risk index of each cell block region is compared with the maximum health risk index. Block regions with a health risk index greater than the maximum health risk index are eliminated. From the remaining block regions, the block region with the smallest health risk index value is selected as the target output region. A risk optimization configuration report is generated. This report records the row and column position number of the target output region in the battery to be configured, the specific value of the health risk index of the region, the safety margin percentage of the region relative to the maximum health risk index, and a restructuring suggestion based on the region.

[0065] Furthermore, S1 collects historical operational images of each of the multiple retired power batteries, specifically:

[0066] Surface image data and internal structure image data of each retired power battery at multiple time points during its historical service phase are acquired. The surface image data and internal structure image data of the same retired power battery at the same time point are spatiotemporally aligned to form aligned image pairs of each retired power battery at multiple time points. The historical operation image sequence of each retired power battery is constructed based on the aligned image pairs.

[0067] In this embodiment, historical operational images of multiple retired power batteries are collected. Specifically, for each retired power battery, all monitoring image data recorded during its service life is retrieved from the historical data repository of the battery management system. The monitoring image data is divided into two categories: the first category is surface image data, which is taken from a top-down angle by a visible light camera installed above the battery module. The shooting frequency is set to trigger once after each charge-discharge cycle, and the exposure time is fixed at 5 milliseconds. The second category is internal structure image data, which is taken synchronously at the same shooting time by the battery management system through an internal voltage and temperature monitoring sensor array. The collected voltage and temperature distribution heatmaps are used to construct the image. The voltage distribution heatmap has a resolution of 16x16 grid points, with each grid point corresponding to the voltage sample value of a single battery cell. The temperature distribution heatmap also has a resolution of 16x16 grid points, with each grid point corresponding to the temperature sample value of a single battery cell. Both voltage and temperature sample values ​​are recorded as integer values ​​in millivolts and millidegrees Celsius. For each retired power battery, the surface image data, voltage distribution heatmap, and temperature distribution heatmap of the battery at the same shooting time are used as a raw image triplet and arranged in chronological order according to the shooting time to form the raw image sequence of the battery.

[0068] When performing spatiotemporal alignment processing on surface image data and internal structure image data of the same retired power battery at the same time point, time alignment is performed first, that is, ensuring that the surface image data, voltage distribution heatmap, and temperature distribution heatmap data involved in the alignment are aligned. Figure 3The acquisition timestamps of the data collectors are completely consistent. If only surface image data is available at a certain time point but voltage distribution heatmap or temperature distribution heatmap is missing, all data at that time point is discarded. If surface image data is missing at a certain time point but voltage distribution heatmap and temperature distribution heatmap exist, all data at that time point is also discarded. This ensures that all three types of data exist simultaneously at each time point after alignment. After time alignment, spatial alignment is performed. The steps for spatial alignment are as follows: establish a spatial mapping relationship between the geometric center coordinates of each battery cell identified in the surface image data and the coordinates of each grid point in the voltage distribution heatmap. At the same time, establish the same spatial mapping relationship between the geometric center coordinates and the coordinates of each grid point in the temperature distribution heatmap. The mapping relationship is established based on the physical arrangement number of the battery cell. That is, the cell in the i-th row and j-th column of the surface image corresponds to the grid point in the i-th row and j-th column of the voltage distribution heat map, and also corresponds to the grid point in the i-th row and j-th column of the temperature distribution heat map. After the spatial mapping relationship is established, the pixel value of the rectangular area where each cell is located in the surface image data, the voltage millivolt value of the corresponding grid point in the voltage distribution heat map, and the temperature millidegree Celsius value of the corresponding grid point in the temperature distribution heat map are associated together to form an aligned image pair of the cell at that time node. All aligned image pairs of cells at the same time node are organized according to the cell arrangement order to form a complete aligned image pair group at that time node. The complete aligned image pair group contains the surface visual information and internal electrothermal state information of each cell of the retired power battery at that moment.

[0069] When constructing the historical operation image sequence of each retired power battery using the aligned image pairs as units, the complete aligned image pairs of the same retired power battery at different time points are arranged in the order of timestamps. The time interval between two adjacent complete aligned image pairs is the duration of one charge-discharge cycle for the retired power battery. If the time interval between a certain adjacent time point is greater than twice the average cycle duration, an empty identifier is inserted at that point to indicate that the battery is in a long-term static interval. After the arrangement is completed, each complete aligned image pair is assigned a global cycle number. This number starts counting from zero when the battery is first powered on, and increments by one after each charge-discharge cycle is completed, until the battery is retired. The complete aligned image pairs carrying global cycle numbers are aggregated to form the historical operating image sequence of the retired power battery. The data storage structure in this sequence is as follows: each element of the sequence is a complete aligned image pair, which contains aligned image pairs of all individual cells at that time point. Each aligned image pair contains pixel matrix data of the rectangular area of ​​the surface image, integer values ​​of voltage in millivolts and temperature in millidegrees, all of which are stored in the sequence element in numerical data format. After performing the above acquisition and sequence construction operations on all retired power batteries, multiple independent historical operating image sequences of retired power batteries are obtained. The length of each sequence varies depending on the service time and number of cycles of each battery.

[0070] Furthermore, S1 performs health clustering annotation processing on each of the historical operating images to generate a set of cell-annotated historical images, specifically:

[0071] Extract the cell outline boundaries from each of the historical operation images, and cut each historical operation image into several cell sub-images based on the cell outline boundaries.

[0072] Image feature vectors characterizing the aging degree of each cell sub-image are extracted. The image feature vectors are input into the health clustering space for clustering operations. Based on the clustering operation results, the health level label corresponding to each cell sub-image is determined. The cell sub-images carrying the health level label are then aggregated into a set of labeled historical images of the cells.

[0073] In this embodiment, health clustering and labeling processing is performed on each of the historical operation images to generate a set of labeled historical images of the battery cells. Specifically, complete aligned image pairs at each time point are extracted sequentially from the historical operation image sequence. For each extracted complete aligned image pair, the surface image data is first extracted, and a battery cell contour boundary recognition operation is performed on the surface image data. The recognition operation adopts a combination of edge intensity detection and connected component analysis. First, the brightness gradient magnitude of each pixel position in the surface image in the horizontal and vertical directions is calculated. Pixels with brightness gradient magnitudes greater than a preset edge threshold are marked as candidate edge pixels. Then, eight-connected component connection analysis is performed on all candidate edge pixels to connect the interconnected components. Candidate edge pixels are grouped into a connected region. Then, based on the bounding rectangle size and aspect ratio of the connected region, connected regions conforming to the geometric size range of a single battery cell are selected as valid battery cell outlines. The coordinates of the bounding rectangle of the valid battery cell outline are recorded as the bounding box of that battery cell in the surface image. Based on the coordinates of this bounding box, a rectangular area defined by the bounding box is cropped from the surface image data as the surface sub-image of that battery cell. Simultaneously, the millivolt voltage value corresponding to the row and column position of that battery cell is extracted from the voltage distribution heatmap, and the millidegree temperature value corresponding to the row and column position of that battery cell is extracted from the temperature distribution heatmap. The surface sub-image, millivolt voltage value, and millidegree temperature value of that battery cell are combined into a single battery cell data unit. All row and column positions in the fully aligned image pair are traversed to obtain the battery cell data units corresponding to each battery cell at that time point.

[0074] Image feature vectors characterizing the aging degree of each individual battery cell are extracted. The extraction operation is performed on the surface sub-image of each individual battery cell. First, the surface sub-image is divided into multiple non-overlapping local blocks, each occupying a fixed number of pixels horizontally and vertically. For each local block, the mean values ​​of the red, green, and blue channels are calculated, along with their standard deviations. These six statistical values ​​are used as the local texture features of the local block. The local texture features of all local blocks are concatenated according to their row and column order in the surface sub-image to form the texture feature vector of the surface sub-image. This texture feature vector is then concatenated with the millivolt voltage and millidegree Celsius temperature values ​​of the individual battery cell to form the complete image feature vector of the individual battery cell at that time point. The dimension of this complete image feature vector is equal to the dimension of the texture feature vector plus two.

[0075] The image feature vector is input into the health clustering space for clustering operations. The health clustering space is a pre-defined multi-dimensional feature space with the same dimension as the complete image feature vector. Several health clustering centers are pre-defined within this space, each corresponding to a health level. The health levels are divided into five levels from high to low, with level one representing the mildest cell aging and level five representing the most severe cell aging. The clustering operation process is as follows: calculate the Euclidean distance between the complete image feature vector of each individual cell data unit and each health clustering center; find the health clustering center with the smallest distance; and then select the health clustering center. The health level corresponding to the Kangdu cluster center is assigned to the individual battery cell data unit as its health level label. At the same time, based on the correspondence between the global cycle number corresponding to the individual battery cell data unit and the health level label, the confidence level of the health level label is determined. If the health level label of the same battery cell changes unidirectionally from low to high level at three consecutive time nodes, the confidence level of these labels is marked as high confidence. If the health level label repeatedly jumps between adjacent time nodes, the confidence level of these labels is marked as low confidence, and the image feature vector of the battery cell at the corresponding time node is re-clustered and reassigned until the jump is eliminated.

[0076] Each battery cell data unit carrying the health level label is categorized according to its row and column position. All battery cell data units belonging to the same row and column position are arranged in ascending order of global cycle number to form the historical time series data chain of that battery cell. All historical time series data chains of battery cells are aggregated to form a battery cell labeled historical image set. Each element of this set is a battery cell data unit. Each battery cell data unit contains at least seven fields: row and column position number, global cycle number, surface sub-image pixel value matrix, voltage millivolt value, temperature millidegree Celsius value, health level label, and confidence level mark.

[0077] Furthermore, referring to Figure 2 Specifically, S2 is:

[0078] Obtain the original acquisition timestamps corresponding to each image in the battery cell labeled historical image set, and determine the key sampling time sequence based on the distribution density of each original acquisition timestamp on the time axis;

[0079] Using the key sampling time sequence as an index, the labeled images that match each key sampling time are extracted from the set of labeled historical images of the battery cell to form a subset of key sampling images;

[0080] Spatial resolution alignment processing is performed on each image in the key sampled image subset to give each image a uniform spatial pixel scale;

[0081] Temporal noise filtering is performed on each image after spatial resolution alignment to suppress random temporal fluctuation components introduced during image acquisition.

[0082] The images, after spatial resolution alignment and temporal noise filtering, are compiled into a set of historical images for battery cell training.

[0083] In this embodiment, the global cycle number field of all individual cell data units is first read from the cell-annotated historical image set. For each individual cell, its global cycle number at all time points is arranged into an increasing sequence. The difference between two adjacent global cycle numbers in this increasing sequence is calculated to obtain the cycle number increment sequence of that individual cell. An increment threshold of three cycles is set, and each increment value in the cycle number increment sequence is traversed. When the increment value at a certain position is greater than three cycles, the time point corresponding to the increment value is marked as the candidate key sampling time of that individual cell. Simultaneously, for the health level label sequence of the battery cell at all time points, it is checked whether the level labels between adjacent time points have changed. If the health level label between adjacent time points changes from a certain level to a higher level, the time point where the change occurs is also marked as a candidate key sampling moment for the battery cell. In addition, differential operation is performed on the voltage millivolt value sequence of the battery cell at all time points to calculate the voltage change between adjacent time points. When the absolute value of the voltage change at a certain position exceeds 200 millivolts per string, the time point corresponding to that position is marked as a candidate key sampling moment. Combining the above three marking conditions, the set of all candidate key sampling moments for the battery cell is obtained.

[0084] The process involves extracting labeled images matching each key sampling moment from the cell-labeled historical image set. Specifically, for each cell, all its candidate key sampling moments are arranged in chronological order. For the arranged candidate key sampling moment sequence, the difference in global cycle number between two adjacent candidate key sampling moments is compared one by one. If the difference is less than a preset minimum interval threshold (set to five cycles), the candidate key sampling moment with the lower health level label or lower confidence level is deleted, and the other is retained. The remaining candidate key sampling moments after this redundancy removal process constitute the final key sampling moment sequence for the cell. Based on the global cycle number corresponding to each moment in the final key sampling moment sequence, all field data of the cell under that global cycle number are retrieved from the cell-labeled historical image set, including the surface sub-image pixel value matrix, voltage millivolt value, and temperature millidegree Celsius value. These retrieved data are encapsulated into key sampling image units. The above extraction operation is performed on all cell cells, and all extracted key sampling image units are summarized to form a key sampling image subset.

[0085] Spatial resolution alignment processing is performed on each image in the key sampled image subset. Specifically, for each surface sub-image pixel value matrix in the key sampled image subset, the current horizontal and vertical pixel counts are obtained. Based on the preset target horizontal pixel count of 1280 and target vertical pixel count of 720, the horizontal scaling ratio and vertical scaling ratio are calculated. The horizontal scaling ratio is equal to the target horizontal pixel count divided by the current horizontal pixel count, and the vertical scaling ratio is equal to the target vertical pixel count divided by the current vertical pixel count. The surface sub-images are scaled using bilinear interpolation. That is, for each target pixel position in the target image, it is back-mapped to the sub-pixel position in the original image according to the scaling ratio. The pixel values ​​of the four nearest integer pixels around the sub-pixel position are selected and weighted by distance to obtain the target pixel value. When weighting the average, the closer the original pixel is, the greater the weight, and the farther the original pixel is, the smaller the weight. After the above bilinear interpolation scaling, the spatial resolution of all surface sub-images is unified to 1280 pixels by 720 pixels, and the red, green, and blue channel values ​​of each pixel still remain integers in the range of 0 to 255.

[0086] Temporal noise filtering is performed on each spatially aligned image. Specifically, the following steps are taken: The spatially aligned surface sub-images are grouped according to the cell unit identifier. Surface sub-images belonging to the same cell unit are arranged in ascending order of their global loop number to form an image time series. For each frame of the surface sub-image in this image time series, using that frame as the center frame, the two frames preceding and following it are selected as reference frames. If the frame is located at the beginning or end of the sequence, resulting in insufficient reference frames, it is copied to make a total of five frames. For each pixel position in the center frame, a sequence of five values ​​is formed by combining the red channel values ​​of that pixel position in the center frame and the four reference frames. The median of this sequence is calculated, and the median is used as the median... As the new red channel value for this pixel position in the center frame, the green and blue channel values ​​are processed in the same way to calculate their respective medians. After all pixel positions have undergone the above median filtering process, the temporal denoised surface sub-image of this frame is obtained. The channel value of each pixel position in this denoised image is the median of the original five temporal channel values, thereby effectively suppressing the temporal noise introduced by random fluctuations between frames during image acquisition. All surface sub-images after spatial resolution alignment and temporal noise filtering, together with their corresponding millivolt voltage values, millidegree Celsius temperature values, health level labels, row and column position numbers, and global cycle numbers, are reorganized according to the two-dimensional index of the individual cell and the global cycle number, and summarized into a set of historical images for cell training.

[0087] Furthermore, referring to Figure 3 S3 specifically refers to:

[0088] S31, the set of battery cell training history images is divided into several training batches, and each training batch contains an equal number of images.

[0089] S32, input the images of the current training batch into the cell tiered configuration clustering network, the cell tiered configuration clustering network extracts aging feature representations for each image, and assigns each image to the corresponding aging gradient clustering cluster according to the similarity measure between the aging feature representations;

[0090] S33, After completing the allocation of the current training batch, update the cluster center features of each cluster in the cell tiered configuration clustering network according to the allocation results;

[0091] S34, use the updated cluster center features as the initial cluster centers for the next training batch, repeat the image allocation and cluster center update operations until all training batches are processed, and output the cell aging gradient clustering configuration model.

[0092] In this embodiment, the total number of all image units is read from the battery cell training history image set. This total number is divided by a preset batch size, which is set to 256 image units. If the division result is not an integer, it is rounded up to obtain the total number of training batches. All image units are randomly divided into a corresponding number of training batches according to the total number of batches. Each training batch contains an equal number of image units, which is 256. If the last batch has less than 256, it is supplemented with randomly sampled image units until it reaches 256. All training batches are arranged in the order after division to form a batch queue.

[0093] When the images of the current training batch are input into the cell tiered configuration clustering network, the network first extracts the pixel value matrix of the surface sub-image from each image unit in the batch, performs global statistical calculations on the pixel value matrix, and calculates the global mean, global mean, and global mean of the red channel, the green channel, and the blue channel, as well as the global standard deviation of the red channel, the green channel, and the blue channel, respectively. These six statistical measures are used as the global statistical features of the image unit. At the same time, the network normalizes and adjusts the millivolt voltage and millidegree Celsius temperature values ​​in the image unit, and then concatenates them with the above global statistical features to form a complete feature description vector. Each image unit corresponds to one feature description vector. After the 256 image units in the current batch generate their corresponding feature description vectors, all feature description vectors are combined into a feature description vector set.

[0094] Based on the similarity measure between the feature description vectors, each image is assigned to a corresponding aging gradient cluster. Specifically, before the current batch is input, the network maintains a set of current aging gradient clusters. Each cluster has a cluster center feature vector, and the dimension of the cluster center feature vector is the same as the dimension of the feature description vector. For the feature description vector of each image unit in the current batch, the Euclidean distance between the feature description vector and each cluster center feature vector is calculated. The Euclidean distance is calculated by summing the squares of the differences between the two vectors in each dimension and then taking the square root. Among all the Euclidean distances, the smallest distance is found, and the cluster corresponding to this distance is determined as the cluster to which the image unit belongs. The image unit is assigned to this cluster, and the identifier of the image unit and the distance value between its feature description vector and the cluster center feature vector of the assigned cluster are recorded.

[0095] After the current training batch is allocated, the cluster center features of each cluster in the cell tiered configuration clustering network are updated according to the allocation results. Specifically, for each cluster, the feature description vectors of all image units currently allocated to the cluster are collected, and the arithmetic mean of these feature description vectors is taken in the same dimension. That is, for each dimension of the feature description vector, the sum of the values ​​of all feature description vectors in the cluster in that dimension is calculated and divided by the number of image units in the cluster. The average values ​​of each dimension are used to form a new cluster center feature vector, and the original cluster center feature vector of the cluster is replaced with the new cluster center feature vector. If a cluster is not allocated any image units in the current batch, the cluster center feature vector of the cluster remains unchanged. After all clusters have completed the cluster center feature vector update, the change of each cluster center feature vector before and after the update is recorded. The change is defined as the Euclidean distance between the updated cluster center feature vector and the unupdated cluster center feature vector.

[0096] The updated cluster center features are used as the initial cluster centers for the next training batch. The image allocation and cluster center update operations are repeated: the next training batch is loaded from the batch queue, the feature description vectors of the image units in that batch are distance-calculated and allocated to the updated cluster center feature vectors, and then the cluster center feature vectors are updated again based on the allocation results. This process is repeated for all training batches. After all batches in the batch queue have been processed, the cumulative change in the cluster center feature vectors of all clusters during this complete traversal is calculated. This cumulative change is the minimum change in the cluster center feature vectors of each cluster during this round. The sum of the Euclidean distances between the final cluster center feature vector and the final cluster center feature vector of the previous round; if the cumulative change is greater than the preset convergence threshold, which is set to 0.01 times the feature vector dimension, then the division order of all image units is shuffled again, a new batch queue is generated, and the next round of traversal iteration is started; if the cumulative change is less than or equal to the convergence threshold, then the iteration stops, and the cluster center feature vectors of each current cluster, the list of image unit identifiers contained in each cluster, and the statistical value of the number of image units in each cluster are solidified and stored as parameter data of the cell aging gradient clustering configuration model.

[0097] Furthermore, S32 specifically includes:

[0098] The cell tiered configuration clustering network performs multi-layer feature abstraction operations on each image, extracting multi-scale aging feature representations from each image, ranging from local texture features to global degradation distribution features;

[0099] Calculate the distance metric between the multi-scale aging feature representation of each image and the cluster center features of each current cluster;

[0100] Based on the distance metric, the target cluster with the smallest distance metric to each image is determined, and each image is assigned to the corresponding target cluster.

[0101] In this embodiment, the images of the current training batch are input into the cell tiered configuration clustering network. The cell tiered configuration clustering network extracts aging feature representations for each image and assigns each image to a corresponding aging gradient cluster based on the similarity measure between the aging feature representations. Specifically, each image unit in the current training batch includes a surface sub-image pixel value matrix, a voltage millivolt value, and a temperature millidegree Celsius value. The cell tiered configuration clustering network performs a multi-layer feature abstraction operation on this image unit. The first layer of the multi-layer feature abstraction operation is to perform local neighborhood feature extraction on the surface sub-image pixel value matrix. Extraction is performed by taking a 5-pixel multiplied 5-pixel neighborhood window centered on each pixel position in the pixel value matrix. The difference between the maximum and minimum values ​​of the red channel pixel values ​​within this neighborhood window is calculated as the red local contrast. The green and blue channels are calculated in the same way, and the local contrast values ​​of the three channels are combined to form the local texture response value of the pixel position. After traversing and calculating all pixel positions in the pixel value matrix, the local texture response map of the surface sub-image is obtained. The size of the response map is the same as the size of the original pixel value matrix.

[0102] The second layer of the multi-layer feature abstraction operation divides the local texture response map output from the first layer into multiple non-overlapping regions. Each region occupies 32 pixels horizontally and vertically. The mean and variance of the local texture response values ​​for all pixels within each region are calculated. These mean and variance values ​​are then concatenated according to the row and column order of the regions to form the mid-level structural feature vector of the surface sub-image. The third layer of the multi-layer feature abstraction operation performs global aggregation on the mid-level structural feature vector output from the second layer, that is, it aggregates the values ​​in all dimensions of the mid-level structural feature vector according to... The images are sorted from largest to smallest. The average of the first 10% of the sorted values ​​is taken as the global activity feature, and the average of the last 10% of the sorted values ​​is taken as the global stationarity feature. The global activity feature and the global stationarity feature are combined to form the high-level degradation distribution feature of the surface sub-image. The high-level degradation distribution feature is then concatenated with the voltage millivolt value, temperature millidegree Celsius value, health level label, and current global cycle number in the image unit to form the multi-scale aging feature representation vector of the image unit. The dimension of the multi-scale aging feature representation vector is equal to the dimension of the high-level degradation distribution feature plus four.

[0103] When calculating the distance metric between the multi-scale aging feature representation of each image and the cluster center features of each current cluster, a weighted Euclidean distance is used. The weighted Euclidean distance is calculated as follows: for each dimension of the feature representation vector and the cluster center feature vector, the difference in that dimension is first calculated, the difference is squared and then multiplied by the preset weight coefficient corresponding to that dimension, the weighted square values ​​of all dimensions are summed and then the square root is taken. The preset weight coefficient is pre-set according to the importance of each dimension in distinguishing different aging degrees. The higher the importance of the dimension, the larger the weight coefficient, and the lower the importance of the dimension, the smaller the weight coefficient. For each image unit in the current batch, the weighted Euclidean distance between its multi-scale aging feature representation vector and the cluster center feature vector of all current clusters is calculated to obtain a set of distance values. Each distance value in this set of distance values ​​corresponds to a cluster.

[0104] Based on the distance metric values, the target cluster with the smallest distance metric value to each image is determined. Specifically, for each image unit in the current batch, the smallest distance value is found from its corresponding set of distance values, and the cluster corresponding to this distance value is determined as the target cluster. The image unit is then assigned to this target cluster. If multiple clusters have the same minimum distance value to a certain image unit, the cluster with the fewest currently assigned image units is selected as the target cluster for that image unit to maintain a balance in the number of samples among the clusters. After the assignment is completed, the distance value between each image unit and its target cluster, as well as the sequence number of the image unit within the target cluster, are recorded.

[0105] Furthermore, in step S4, the cell aging gradient clustering configuration model is used to perform gradient clustering block processing on the current operating image of the retired power battery to be configured, and the degradation state detection results of each cell block image to be configured are obtained, specifically:

[0106] Obtain the current operating image of the retired power battery to be configured at the current moment, and cut the current operating image into several cell block images to be configured according to the physical arrangement structure of the cells.

[0107] Each of the cell block images to be configured is input into the cell aging gradient clustering configuration model one by one, and the cell aging gradient clustering configuration model determines the gradient clustering position of each cell block image to be configured in the aging gradient space.

[0108] Based on the gradient clustering position of each cell block image to be configured, the degradation state description parameters corresponding to each gradient clustering position are retrieved from the cell aging gradient clustering configuration model, and the degradation state description parameters are output as the degradation state detection results of each cell block image to be configured.

[0109] In this embodiment, the cell aging gradient clustering configuration model is used to perform gradient clustering block processing on the current operating image of the decommissioned power battery to be configured, so as to obtain the degradation state detection results of each cell block image. Specifically, the current operating image of the decommissioned power battery to be configured at the current moment is obtained. The acquisition method is to place the decommissioned power battery to be configured in the same shooting environment as the historical operating image acquisition, and use the same type of visible light camera with the same top angle, the same exposure time of five milliseconds, and the same white balance color temperature of 5500 Kelvin to shoot a frame of surface image. At the same time, the voltage distribution heat map and the temperature distribution heat map are collected by the voltage monitoring sensor array and the temperature monitoring sensor array respectively at the same moment. The surface image, voltage distribution heat map and temperature distribution heat map are combined into a current operating image data package. The spatial resolution and data format of the current operating image data package are consistent with those of the historical operating images.

[0110] The current running image is divided into several cell block images to be configured according to the physical arrangement structure of the cells. The cutting operation is based on the physical arrangement structure parameters of the cells, which record the total number of rows and columns of the cells in the retired power battery to be configured, as well as the horizontal pixel width and vertical pixel height occupied by each cell in the surface image. According to the total number of rows and columns and the pixel width and pixel height of each cell, a corresponding number of rectangular blocks are cropped from the surface image of the current running image. The horizontal pixel width and vertical pixel height of each rectangular block are fixed, and the blocks do not overlap with each other. After all blocks are stitched together, they completely cover the entire surface image area. For each block cropped, the pixel value matrix of the block is packaged together with the voltage millivolt value and temperature millidegree Celsius value extracted from the current running image data packet that correspond to the row and column position of the block to form a cell block image to be configured. After all blocks are processed, a number of cell block images to be configured are obtained that are equal to the total number of cells.

[0111] Each of the cell block images to be configured is input into the cell aging gradient clustering configuration model. For each input cell block image, the model extracts the multi-scale aging feature representation vector of the block image. The extraction method is exactly the same as the method of extracting feature representations from training images during the model training process, that is, performing local neighborhood contrast calculation, regional block mean and variance statistics, global activity and stability aggregation in sequence, and then concatenating the voltage millivolt value and temperature millidegree Celsius value of the block image to form the feature vector of the cell block image to be configured. The model calculates the weighted Euclidean distance between the feature vector and the cluster center feature vectors of each cluster stored in itself, finds the cluster with the smallest distance, and uses the cluster number of the cluster as the gradient cluster position of the cell block image to be configured in the aging gradient space. The gradient cluster position is an integer value, and the value range is from one to the total number of clusters included in the model.

[0112] Based on the gradient clustering position of each cell block image to be configured, degradation state description parameters corresponding to each gradient clustering position are retrieved from the cell aging gradient clustering configuration model. These degradation state description parameters are stored in the model's parameter data. Each cluster corresponds to a set of degradation state description parameters, including the lower limit of the capacity decay interval, the upper limit of the capacity decay interval, the lower limit of the internal resistance increase interval, the upper limit of the internal resistance increase interval, the lower limit of the self-discharge rate interval, and the upper limit of the self-discharge rate interval. The lower and upper limits of the capacity decay interval are expressed as a percentage of the initial capacity, the lower and upper limits of the internal resistance increase interval are expressed as a percentage of the initial internal resistance, and the lower and upper limits of the self-discharge rate interval are expressed as millivolts of voltage drop per day. The retrieved degradation state description parameters are output as the degradation state detection result for each cell block image to be configured. A set of six values ​​representing the degradation state detection result is output for each cell block image to be configured.

[0113] Furthermore, in S4, the degradation state detection results of each of the cell blocks to be configured are fused and integrated to calculate the health risk index of the retired power battery to be configured, specifically:

[0114] Obtain the degradation state detection results of each cell block image to be configured, wherein the degradation state detection results include the state description values ​​of each cell block in multiple degradation dimensions;

[0115] Based on the physical location weight and electrical connection topology weight of each cell block in the decommissioned power battery to be configured, the state description values ​​of each cell block in multiple degradation dimensions are weighted and aggregated to obtain the overall state aggregate value of the decommissioned power battery to be configured in each degradation dimension.

[0116] The overall state aggregation value on each degradation dimension is mapped to the health risk assessment space, and the health risk index of the power battery to be configured for retirement is calculated through the risk contribution weight corresponding to each degradation dimension in the health risk assessment space.

[0117] In this embodiment, the degradation state detection results of each cell block image to be configured are fused and integrated to calculate the health risk index of the retired power battery to be configured. Specifically, the degradation state detection results of each cell block image to be configured are read one by one from the model output. The degradation state detection results include six values: lower and upper limits of capacity decay range, lower and upper limits of internal resistance growth range, and lower and upper limits of self-discharge rate range. For each cell block image to be configured, its capacity decay... The arithmetic mean of the lower and upper limits of the subtraction interval is used to obtain the representative value of capacity decay for the cell block. The arithmetic mean of the lower and upper limits of its internal resistance growth interval is used to obtain the representative value of internal resistance growth for the cell block. The arithmetic mean of the lower and upper limits of its self-discharge rate interval is used to obtain the representative value of self-discharge rate for the cell block. After the above averaging process, the degradation state detection results of each cell block are obtained, and each block obtains a state description value in three degradation dimensions, namely, the representative value of capacity decay, the representative value of internal resistance growth, and the representative value of self-discharge rate.

[0118] Based on the physical location weight and electrical connection topology weight of each cell block in the decommissioned power battery to be configured, a weighted aggregation operation is performed on the state description values ​​of each cell block in multiple degradation dimensions. Specifically, the physical location weight is determined according to the row and column position of the cell block in the battery module. If the block is located in any of the four corner regions of the module, a physical location weight of 1.2 is assigned; if the block is located in the edge region of the module but not in the corner region, a physical location weight of 1.0 is assigned; if the block is located in the inner center region of the module, a physical location weight of 1.0 is assigned. The weight of the electrical connection topology is determined by the position of the cell block in the battery series-parallel topology. If the block is located in the first part of the series branch near the positive output terminal or the last part near the negative output terminal, the electrical connection topology weight is 1.3. If the block is located at the parallel node, the electrical connection topology weight is 1.1. If the block is located in the middle of the series branch and not at the parallel node, the electrical connection topology weight is 0.9. For each cell block, the physical location weight is multiplied by the electrical connection topology weight to obtain the comprehensive weighting coefficient of the block.

[0119] Weighted aggregation operations are performed on each degradation dimension. For the capacity decay dimension, the representative value of capacity decay for each cell block is multiplied by the comprehensive weighting coefficient of that block and then summed to obtain the weighted sum of capacity decay. The comprehensive weighting coefficients of all cell blocks are summed to obtain the total comprehensive weighting coefficient. The weighted sum of capacity decay is divided by the total comprehensive weighting coefficient to obtain the overall capacity decay aggregated value. For the internal resistance growth dimension, the representative value of internal resistance growth for each cell block is multiplied by the comprehensive weighting coefficient of that block and then summed to obtain the weighted sum of internal resistance growth. The weighted sum of internal resistance growth is divided by the total comprehensive weighting coefficient to obtain the overall internal resistance growth aggregated value. For the self-discharge rate dimension, the representative value of self-discharge rate for each cell block is multiplied by the comprehensive weighting coefficient of that block and then summed to obtain the weighted sum of self-discharge rate. The weighted sum of self-discharge rate is divided by the total comprehensive weighting coefficient to obtain the overall self-discharge rate aggregated value.

[0120] The overall state aggregation value across each degradation dimension is mapped to a health risk assessment space, which is a three-dimensional risk assessment coordinate system. The three axes correspond to capacity decay risk, internal resistance growth risk, and self-discharge rate risk, respectively. For the overall capacity decay aggregation value, the corresponding capacity risk component value is determined based on its position on the capacity decay risk mapping curve. When the overall capacity decay aggregation value is in the range of 0 to 10%, the corresponding capacity risk component increases linearly from 0 to 20; when it is in the range of 10% to 20%, the corresponding capacity risk component increases linearly from 20 to 60; and when it is above 20%, the corresponding capacity risk component increases linearly from 60 to 100. For the overall internal resistance growth aggregate value, the corresponding internal resistance risk component is determined based on the position of the aggregate value on the internal resistance growth risk mapping curve. For the overall self-discharge rate aggregate value, the corresponding self-discharge rate risk component is determined based on the position of the aggregate value on the self-discharge rate risk mapping curve. The capacity risk component, internal resistance risk component, and self-discharge rate risk component are added together to obtain the health risk index of the retired power battery to be configured. The minimum value of the health risk index is 0, and the maximum value is 100. The larger the value, the higher the overall health risk.

[0121] Furthermore, S5 specifically includes:

[0122] Obtain the cycle length parameter and working condition intensity parameter corresponding to the expected service cycle, and determine the health risk index threshold based on the cycle length parameter and working condition intensity parameter;

[0123] The health risk index of each cell block region in the power battery to be configured is compared with the health risk index threshold, and candidate cell block regions with health risk indices lower than the health risk index threshold are selected.

[0124] Select the cell block region with the lowest health risk index from the candidate cell block regions as the target configuration region, and generate a risk optimization configuration report that includes the spatial location information of the target configuration region, the health risk index value, and the matching degree evaluation information with the expected service life.

[0125] In this embodiment, based on the matching relationship between the health risk index and the expected service life, the cell area with the lowest health risk index is selected from each cell block area of ​​the power battery to be configured for retirement as the output, and a risk optimization configuration report is generated. In specific execution, the expected service life input by the user is obtained. The expected service life is represented by the number of cycles that the battery needs to complete in the tiered utilization scenario. The expected number of cycles input by the user through the human-computer interaction interface is a positive integer, and the range of the positive integer is set to any integer value between 300 and 3000. Based on the expected number of cycles, the maximum allowable health risk index is found from a pre-defined cycle-risk comparison table. The cycle-risk comparison table is established by pre-calibrating the maximum allowable health risk index of the battery under different cycle requirements through experiments. For example, the maximum allowable health risk index is 75 when the expected number of cycles is 300, 65 when the expected number of cycles is 500, 55 when the expected number of cycles is 800, 45 when the expected number of cycles is 1000, 35 when the expected number of cycles is 1500, 25 when the expected number of cycles is 2000, and 0=10 when the expected number of cycles is 3000. The maximum allowable health risk index is calculated using linear interpolation for the expected number of cycles between the above calibration points.

[0126] The health risk index of each cell block in the decommissioned power battery to be configured is calculated one by one. The calculation method for the health risk index of each cell block is as follows: Take the lower limit and upper limit of the capacity decay range from the degradation state detection results of the cell block, and take their arithmetic mean to obtain the capacity decay value of the block. Map this capacity decay value to a capacity risk component according to the capacity decay risk mapping curve. Take the lower limit and upper limit of the internal resistance growth range, and take their arithmetic mean to obtain the internal resistance growth value. Map this to an internal resistance risk component according to the internal resistance growth risk mapping curve. Take the lower limit and upper limit of the self-discharge rate range, and take their arithmetic mean to obtain the internal resistance growth value. The self-discharge rate value is obtained by averaging the values. It is then mapped to a self-discharge rate risk component according to the self-discharge rate risk mapping curve. The three risk components are added together to obtain the health risk index of the cell block region. After the health risk index of each cell block region is calculated, the health risk index of each block region is compared with the maximum allowable health risk index. If the health risk index of a block region is greater than the maximum allowable health risk index, the block region is marked as an unqualified region. If the health risk index of a block region is less than or equal to the maximum allowable health risk index, the block region is marked as a candidate region.

[0127] From all cell blocks marked as candidate regions, select the cell block with the lowest health risk index value as the target configuration region. Record the row and column position number of the target configuration region, the specific value of the health risk index of the region, and the safety margin percentage of the region relative to the maximum allowable health risk index. The safety margin percentage is calculated by subtracting the health risk index of the target configuration region from the maximum allowable health risk index, dividing by the maximum allowable health risk index, and then multiplying by 100. Generate a risk optimization configuration report, which records the following content in a combination of tables and text descriptions: the row and column position number of the target configuration region, the specific value of the health risk index of the target configuration region (rounded to one decimal place), the safety margin percentage of the target configuration region (rounded to an integer), the representative values ​​of capacity decay, internal resistance growth, and self-discharge rate of the cell corresponding to the target configuration region, and a reorganization suggestion based on the target configuration region. The reorganization suggestion specifies that when constructing a reorganized battery module centered on the target configuration region, priority should be given to combining the cells in this region with other cells whose health risk index is also lower than the maximum allowable health risk index.

[0128] Furthermore, after generating the risk-optimized configuration report, it also includes:

[0129] The risk optimization configuration report is output to the tiered configuration decision terminal. The tiered configuration decision terminal generates a cell reorganization operation instruction for the retired power battery to be configured based on the spatial location information of the target configuration area in the risk optimization configuration report. The cell reorganization operation instruction includes the cell area identifier to be removed and the cell area identifier to be retained.

[0130] In this embodiment, after generating the risk optimization configuration report, the risk optimization configuration report is output to the tiered configuration decision terminal. The tiered configuration decision terminal generates a cell recombination operation instruction for the retired power battery to be configured based on the spatial location information of the target configuration area in the risk optimization configuration report. Specifically, during execution, the risk optimization configuration report is sent to the tiered configuration decision terminal in the form of a structured data file through a network transmission interface. This terminal is an industrial control computer equipped with a display and operation input devices. After receiving the report, the terminal parses out the row and column position numbers of the target configuration area and the list of row and column position numbers of all cell blocks marked as candidate areas.

[0131] The tiered configuration decision terminal queries the physical layout mapping table of the retired power batteries to be configured based on the row and column position numbers of the target configuration area. The physical layout mapping table records the physical coordinates of the individual cells corresponding to each row and column position number in the battery module, including the horizontal and vertical installation coordinates of the cell on the module base plate, as well as the welding point coordinates of the connector terminals of the positive and negative terminals of the cell on the module busbar. After determining the physical coordinates of the individual cells based on the row and column position numbers of the target configuration area, the terminal marks the individual cells as cells to be retained. At the same time, the terminal removes the row and column position numbers of the target configuration area from the list of row and column position numbers of all candidate areas, marks the remaining candidate area cells as cells to be retained, and marks the cell blocks that are not marked as candidate areas and the cell blocks marked as unqualified areas as cells to be cut.

[0132] The process of generating the cell reconfiguration operation command is as follows: The terminal generates a list of cell regions to be removed and a list of cell regions to be retained, based on the markings of the cells to be retained and the cells to be removed. The list of cell regions to be removed contains the row and column position numbers and their physical coordinates of all cells to be removed, while the list of cell regions to be retained contains the row and column position numbers and their physical coordinates of all cells to be retained. The terminal also calculates the minimum rectangular reconfiguration area boundary required to expand outwards from the target configuration area until all cells to be retained are covered, based on the physical adjacency and electrical connection relationships between the cells to be retained. The coordinates of the four corner points of this minimum rectangular reconfiguration area boundary are determined by the physical coordinates of all cells to be retained. The coordinates are obtained by taking the minimum and maximum values ​​of the horizontal and vertical axes. The terminal encapsulates the list of cell regions to be removed, the list of cell regions to be retained, and the boundary coordinates of the minimum rectangular reassembly area into a cell reassembly operation instruction. The instruction is in the format of a binary data packet containing an instruction header, an operation type code, and a parameter data body. The instruction header records the instruction length and verification information. The operation type code is a fixed special code for cell reassembly operations. The parameter data body contains the number of cells to be removed, the row and column position number sequence and physical coordinate sequence of each cell to be removed, the number of cells to be retained, the row and column position number sequence and physical coordinate sequence of each cell to be retained, and the boundary coordinates of the minimum rectangular reassembly area. After the cell reassembly operation instruction is generated, the tiered configuration decision terminal sends the instruction to the execution controller of the automated cell sorting and reassembly equipment. The execution controller controls the robotic arm to sequentially grab and remove the cells to be removed according to the instruction content, and rearranges and welds the cells to be retained according to the boundary of the minimum rectangular reassembly area.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0134] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0135] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for optimizing the tiered configuration of retired power batteries based on the entire life cycle of battery cells, characterized in that, include: S1. Collect historical operating images corresponding to multiple retired power batteries, perform health clustering and labeling processing on each historical operating image, and generate a set of labeled historical images of battery cells. S2, perform key sampling moment decision preprocessing on the cell-annotated historical image set, complete image specification unification processing and signal noise suppression processing, and obtain cell training historical image set; S3, feed the set of battery cell training history images into the battery cell tiered configuration clustering network to perform iterative processing and construct a battery cell aging gradient clustering configuration model. S4. Using the cell aging gradient clustering configuration model, perform gradient clustering block processing on the current operating image of the power battery to be configured for retirement, obtain the degradation state detection results of each cell block image to be configured, perform fusion and integration processing on the degradation state detection results of each cell block image to be configured, and calculate the health risk index of the power battery to be configured for retirement. S5. Based on the matching relationship between the health risk index and the expected service life, select the cell area with the lowest health risk index from each cell block area of ​​the power battery to be configured for retirement as the output, and generate a risk optimization configuration report.

2. The method according to claim 1, characterized in that, S1 collects historical operational images of multiple retired power batteries, specifically: Surface image data and internal structure image data of each retired power battery at multiple time points during its historical service phase are acquired. The surface image data and internal structure image data of the same retired power battery at the same time point are spatiotemporally aligned to form aligned image pairs of each retired power battery at multiple time points. The historical operation image sequence of each retired power battery is constructed based on the aligned image pairs.

3. The method according to claim 2, characterized in that, S1 performs health clustering annotation processing on each of the historical operating images to generate a set of cell-annotated historical images, specifically: Extract the cell outline boundaries from each of the historical operation images, and cut each historical operation image into several cell sub-images based on the cell outline boundaries. Image feature vectors characterizing the aging degree of each cell sub-image are extracted. The image feature vectors are input into the health clustering space for clustering operations. Based on the clustering operation results, the health level label corresponding to each cell sub-image is determined. The cell sub-images carrying the health level label are then aggregated into a set of labeled historical images of the cells.

4. The method according to claim 1, characterized in that, Specifically, S2 is: Obtain the original acquisition timestamps corresponding to each image in the battery cell labeled historical image set, and determine the key sampling time sequence based on the distribution density of each original acquisition timestamp on the time axis; Using the key sampling time sequence as an index, the labeled images that match each key sampling time are extracted from the set of labeled historical images of the battery cell to form a subset of key sampling images; Spatial resolution alignment processing is performed on each image in the key sampled image subset to give each image a uniform spatial pixel scale; Temporal noise filtering is performed on each image after spatial resolution alignment to suppress random temporal fluctuation components introduced during image acquisition. The images, after spatial resolution alignment and temporal noise filtering, are compiled into a set of historical images for battery cell training.

5. The method according to claim 1, characterized in that, Specifically, S3 is: S31, the set of battery cell training history images is divided into several training batches, and each training batch contains an equal number of images. S32, input the images of the current training batch into the cell tiered configuration clustering network, the cell tiered configuration clustering network extracts aging feature representations for each image, and assigns each image to the corresponding aging gradient clustering cluster according to the similarity measure between the aging feature representations; S33, After completing the allocation of the current training batch, update the cluster center features of each cluster in the cell tiered configuration clustering network according to the allocation results; S34, use the updated cluster center features as the initial cluster centers for the next training batch, repeat the image allocation and cluster center update operations until all training batches are processed, and output the cell aging gradient clustering configuration model.

6. The method according to claim 5, characterized in that, Specifically, S32 is as follows: The cell tiered configuration clustering network performs multi-layer feature abstraction operations on each image, extracting multi-scale aging feature representations from each image, ranging from local texture features to global degradation distribution features; Calculate the distance metric between the multi-scale aging feature representation of each image and the cluster center features of each current cluster; Based on the distance metric, the target cluster with the smallest distance metric to each image is determined, and each image is assigned to the corresponding target cluster.

7. The method according to claim 1, characterized in that, In step S4, the cell aging gradient clustering configuration model is used to perform gradient clustering block processing on the current operating image of the retired power battery to be configured, and the degradation state detection results of each cell block image to be configured are obtained, specifically: Obtain the current operating image of the retired power battery to be configured at the current moment, and cut the current operating image into several cell block images to be configured according to the physical arrangement structure of the cells. Each of the cell block images to be configured is input into the cell aging gradient clustering configuration model one by one, and the cell aging gradient clustering configuration model determines the gradient clustering position of each cell block image to be configured in the aging gradient space. Based on the gradient clustering position of each cell block image to be configured, the degradation state description parameters corresponding to each gradient clustering position are retrieved from the cell aging gradient clustering configuration model, and the degradation state description parameters are output as the degradation state detection results of each cell block image to be configured.

8. The method according to claim 7, characterized in that, In step S4, the degradation state detection results of each of the cell blocks to be configured are fused and integrated to calculate the health risk index of the retired power battery to be configured, specifically: Obtain the degradation state detection results of each cell block image to be configured, wherein the degradation state detection results include the state description values ​​of each cell block in multiple degradation dimensions; Based on the physical location weight and electrical connection topology weight of each cell block in the decommissioned power battery to be configured, the state description values ​​of each cell block in multiple degradation dimensions are weighted and aggregated to obtain the overall state aggregate value of the decommissioned power battery to be configured in each degradation dimension. The overall state aggregation value on each degradation dimension is mapped to the health risk assessment space, and the health risk index of the power battery to be configured for retirement is calculated through the risk contribution weight corresponding to each degradation dimension in the health risk assessment space.

9. The method according to claim 1, characterized in that, Specifically, S5 is: Obtain the cycle length parameter and working condition intensity parameter corresponding to the expected service cycle, and determine the health risk index threshold based on the cycle length parameter and working condition intensity parameter; The health risk index of each cell block region in the power battery to be configured is compared with the health risk index threshold, and candidate cell block regions with health risk indices lower than the health risk index threshold are selected. Select the cell block region with the lowest health risk index from the candidate cell block regions as the target configuration region, and generate a risk optimization configuration report that includes the spatial location information of the target configuration region, the health risk index value, and the matching degree evaluation information with the expected service life.

10. The method according to claim 1, characterized in that, After generating the risk optimization configuration report, it also includes: The risk optimization configuration report is output to the tiered configuration decision terminal. The tiered configuration decision terminal generates a cell reorganization operation instruction for the retired power battery to be configured based on the spatial location information of the target configuration area in the risk optimization configuration report. The cell reorganization operation instruction includes the cell area identifier to be removed and the cell area identifier to be retained.

Images