Lithium battery health condition multi-dimensional diagnosis method and system

Abstract

The invention discloses a lithium battery health condition multi-dimensional diagnosis method and system, and relates to the field of lithium battery health diagnos.The lithium battery health condition multi-dimensional diagnosis method comprises the steps that discrete charging data fragments are obtained from a vehicle management system, and intelligent reconstruction of a virtual incremental capacity analysis curve is conducted based on the fragments; a confidence map is synchronously generated to quantify the reliability of the reconstructed data. And then, key health features are extracted from the reconstruction curve, and adaptive weighting is performed on the features by using a confidence map, so that a comprehensive index vector capable of accurately reflecting the real aging state in the battery is generated. And finally, deep diagnosis is carried out based on the index vector, and multi-dimensional accurate diagnosis of the health condition of the battery is realized, so that a multi-dimensional evaluation report of the health condition of the battery is output. Therefore, the interference of poor data quality on the diagnosis precision under the real driving working condition can be effectively overcome, and the practical applicability and reliability of diagnosis are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of lithium battery health diagnosis, and more specifically, to a multi-dimensional diagnostic method and system for the health status of lithium batteries. Background Technology

[0002] With the booming development of the new energy vehicle industry and the large-scale application of energy storage technology, the safety, reliability, and durability of lithium-ion batteries, as the core energy carrier, have become key factors restricting the further development of the industry. During long-term use, lithium batteries experience performance degradation due to complex electrochemical reactions. This dynamic change in State of Health (SOH) directly affects the driving range, charging efficiency, and overall operational safety of electric vehicles. Therefore, to protect user rights, optimize battery management strategies, and enhance the value of secondary use, developing a solution capable of accurately, in real-time, and multi-dimensionally diagnosing the health status of lithium batteries is crucial. Deep insights into battery health can not only provide users with accurate remaining life predictions and maintenance recommendations but also provide key data support for the Battery Management System (BMS), thereby enabling smarter energy scheduling and safety warnings. This has profound significance for promoting the healthy development of the entire industry chain.

[0003] Among existing diagnostic technologies, methods based on electrochemical mechanisms such as incremental capacity analysis (ICA) and differential voltage analysis (DVA) have attracted widespread attention due to their ability to reveal internal aging patterns in batteries. These methods analyze minute changes in voltage curves during charge and discharge to extract features related to aging phenomena such as capacity decay and increased internal resistance, thereby assessing the battery's health status. However, these sophisticated laboratory analytical methods face significant challenges in practical applications, especially for electric vehicles under real-world driving conditions. The accuracy of laboratory-level ICA / DVA analysis highly depends on complete charging data collected at a constant and low current. In the real world, however, user charging behavior is highly random and uncertain. For example, a vehicle owner may only partially charge (e.g., from 30% to 80%), and the charging current may vary due to grid fluctuations, charging station power limitations, or active BMS regulation. This incomplete, non-constant current "data fragment" severely interferes with the feature extraction process of traditional analytical methods, making it difficult to stably and reliably obtain equivalent electrochemical characteristics that accurately reflect the battery's health status, thus significantly reducing the accuracy and reliability of diagnostic results.

[0004] Therefore, there is an urgent need for an optimized multi-dimensional diagnostic method and system for the health status of lithium batteries. Summary of the Invention

[0005] This application is made in order to solve the above-mentioned technical problems.

[0006] According to one aspect of this application, a multi-dimensional diagnostic method for the health status of a lithium battery is provided, comprising:

[0007] Obtain raw BMS data;

[0008] Charging data segments are extracted from the raw BMS data and preprocessed to obtain a normalized set of charging segments.

[0009] Virtual ICA curves and confidence levels are reconstructed and generated for the normalized charging segment set to obtain virtual ICA curves and confidence levels.

[0010] We extract health features based on confidence weighting from the virtual ICA curve and confidence profile to obtain a weighted health index vector;

[0011] Multidimensional health status diagnosis is performed on the weighted health indicator vector to obtain a multidimensional SOH report.

[0012] According to another aspect of this application, a multi-dimensional diagnostic system for the health status of a lithium battery is provided, comprising:

[0013] The raw BMS data acquisition module is used to acquire raw BMS data;

[0014] The charging data preprocessing module is used to extract charging data segments from the raw BMS data and preprocess them to obtain a normalized charging segment set.

[0015] The ICA reconstruction and confidence generation module is used to reconstruct virtual ICA curves and generate confidence scores on a normalized charging segment set to obtain virtual ICA curves and confidence scores.

[0016] The health feature extraction module is used to extract health features from the virtual ICA curve and confidence map based on confidence weighting to obtain a weighted health index vector;

[0017] The multi-dimensional health status diagnosis module is used to perform multi-dimensional health status diagnosis on the weighted health indicator vector to obtain a multi-dimensional SOH report.

[0018] Compared with existing technologies, this application provides a multi-dimensional diagnostic method and system for the health status of lithium batteries. First, it acquires discrete charging data fragments from the vehicle management system and intelligently reconstructs virtual incremental capacity analysis curves based on these fragments, simultaneously generating a confidence graph to quantify the reliability of the reconstructed data. Then, it extracts key health features from the reconstructed curves and adaptively weights these features using the confidence graph, thereby generating a comprehensive index vector that accurately reflects the true aging state of the battery. Finally, it performs deep diagnostics based on this index vector, achieving a multi-dimensional and accurate diagnosis of the battery's health status and outputting a multi-dimensional battery health status assessment report. This effectively overcomes the interference of poor data quality under real-world driving conditions on diagnostic accuracy, significantly improving the practical applicability and reliability of the diagnosis. Attached Figure Description

[0019] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 This is a flowchart of a multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application.

[0021] Figure 2 This is a data flow diagram of a multi-dimensional diagnostic method for the health status of lithium batteries according to an embodiment of this application.

[0022] Figure 3 This is a flowchart of sub-step S2 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application.

[0023] Figure 4 This is a flowchart of sub-step S3 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application.

[0024] Figure 5 This is a flowchart of sub-step S31 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application.

[0025] Figure 6 This is a flowchart of sub-step S32 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application.

[0026] Figure 7 This is a flowchart of sub-step S33 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application.

[0027] Figure 8 This is a flowchart of sub-step S4 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application.

[0028] Figure 9 This is a block diagram of a multi-dimensional diagnostic system for the health status of a lithium battery according to an embodiment of this application. Detailed Implementation

[0029] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0030] To address the problems mentioned above in the background technology, this application proposes a multi-dimensional diagnostic method for the health status of lithium batteries. Figure 1 This is a flowchart of a multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application. Figure 2 This is a data flow diagram of a multi-dimensional diagnostic method for the health status of lithium batteries according to an embodiment of this application. (See attached diagram.) Figure 1 and Figure 2 As shown, the multi-dimensional diagnostic method for the health status of the lithium battery includes the following steps: S1, acquiring raw BMS data; S2, extracting charging data segments from the raw BMS data and preprocessing them to obtain a normalized charging segment set; S3, reconstructing virtual ICA curves and generating confidence scores on the normalized charging segment set to obtain virtual ICA curves and confidence scores; S4, extracting health features based on confidence scores from the virtual ICA curves and confidence scores to obtain a weighted health index vector; S5, performing multi-dimensional health status diagnosis on the weighted health index vector to obtain a multi-dimensional SOH report.

[0031] In the aforementioned multi-dimensional diagnostic method for the health status of lithium batteries, step S1 involves acquiring raw BMS data. It should be understood that lithium battery health status diagnosis relies on core electrochemical parameters and operating condition data during battery operation. The Battery Management System (BMS) is the only carrier that collects and stores this data in real time. The lack of raw BMS data will result in all subsequent diagnostic steps losing data support. Therefore, this application acquires raw BMS data through the communication interface between the BMS and the diagnostic system, providing a complete and tamper-proof data source for subsequent steps such as charging segment extraction and health feature analysis. This ensures that all data used in subsequent diagnostic steps originates from the actual operating state of the battery, avoiding deviations in diagnostic results from the true health status due to data loss or tampering, and laying a data foundation for the accuracy of the entire diagnostic process.

[0032] Specifically, in one possible embodiment, step S1 is implemented as follows: First, a communication connection is established between the diagnostic system and the target lithium battery BMS, using the Controller Area Network (CLAN) bus communication protocol. This protocol is a standard protocol for data transmission in vehicle electronic devices, enabling real-time data interaction between the diagnostic system and the BMS. Second, all operating data within a specified time period is retrieved from the BMS using preset communication commands. The retrieved parameters include timestamps, individual cell voltages, total voltage, charging current, discharging current, battery pack temperature, cell temperature, and SOC value. Finally, all retrieved data is stored in the diagnostic system's database in timestamp order, using a structured database format. The integrity of the stored data is verified; if data loss occurs, a new acquisition request is initiated until complete original BMS data is obtained.

[0033] In the aforementioned multi-dimensional diagnostic method for the health status of lithium batteries, step S2 involves extracting charging data segments from the original BMS data and preprocessing them to obtain a normalized set of charging segments. It should be understood that since the original BMS data contains data from various operating conditions such as battery charging, discharging, and resting, and is subject to interference factors such as sensor noise, charging current fluctuations, and temperature differences, directly using it for health diagnosis would lead to distorted feature extraction and fail to meet the data quality requirements for subsequent virtual ICA curve reconstruction. Therefore, this application filters charging data segments from the original BMS data and obtains charging data that can be directly used for subsequent analysis through preprocessing. This eliminates interference from irrelevant operating condition data, removes the influence of different environments and operating conditions on the data, and ensures the consistency and accuracy of the processed data. This lays the foundation for accurate reconstruction of the virtual ICA curve and reliable extraction of health features, avoiding deviations in diagnostic results from the true health status of the battery due to data quality issues.

[0034] In particular, in one specific embodiment, Figure 3 This is a flowchart of sub-step S2 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application. Figure 3 As shown, step S2 includes: S21, identifying and extracting charging segments from the original BMS data to obtain an original charging segment set; S22, performing data filtering and cumulative capacity calculation on each original charging segment in the original charging segment set to obtain a filtered charging segment set; S23, standardizing the operating conditions of each filtered charging segment in the filtered charging segment set to obtain a normalized charging segment set.

[0035] Specifically, step S21 involves identifying and extracting charging segments from the original BMS data to obtain an original set of charging segments. It should be understood that since the original BMS data is a continuous data stream encompassing all battery operating conditions, non-charging condition data such as discharge and resting conditions are not directly related to the electrochemical characteristics required for lithium battery health diagnosis. Failure to separate these segments would lead to data redundancy in subsequent processing and introduce irrelevant interference, affecting the extraction efficiency and accuracy of charging-related features. Therefore, this application further analyzes the current and voltage change patterns in the original BMS data, identifies and extracts data segments that conform to charging characteristics, thereby obtaining an original set of charging segments containing only the charging process. This allows for rapid screening of charging data directly related to health diagnosis, elimination of redundant interference from irrelevant operating condition data, shortening the time cost of subsequent data preprocessing, and ensuring that subsequent processing objects are all data reflecting the electrochemical state of the battery during the charging stage, laying a data range foundation for subsequent accurate analysis.

[0036] Specifically, in one possible embodiment, step S21 is implemented as follows: First, a charging segment identification threshold is set, including a charging start current threshold, a charging stop current threshold, and a minimum charging duration. Second, the time series of the original BMS data is traversed. When the current jumps from less than the start threshold to greater than the start threshold, it is marked as the charging start point. The tracking continues until the current drops below the stop threshold or the data stream is interrupted, at which point it is marked as the charging end point. Finally, the voltage, current, temperature, and timestamp data between the start and end points are extracted to form the original charging segment. If the segment duration exceeds the minimum threshold, it is added to the original charging segment set, and each segment is checked for integrity to ensure no data loss.

[0037] Specifically, step S22 involves performing data filtering and cumulative capacity calculation on each original charging segment in the original charging segment set to obtain a filtered charging segment set. It should be understood that random noise generated during sensor acquisition in the original charging segments can lead to errors in subsequent incremental capacity calculations. Furthermore, the original segments only contain real-time current and voltage parameters, lacking cumulative capacity data reflecting changes in charge level during charging, which is a core parameter for calculating incremental capacity. Therefore, this application further performs noise filtering on each original charging segment and calculates the cumulative charging capacity based on the relationship between current and time, thereby obtaining a filtered charging segment set with noise elimination and complete parameters. This effectively suppresses the interference of data noise on subsequent feature calculations, supplements the key capacity parameters required for health diagnosis, ensures the accuracy of subsequent incremental capacity calculations, and provides high-quality data support for extracting reliable battery health characteristics.

[0038] Specifically, in one possible embodiment, step S22 is implemented as follows: First, data filtering is performed using a Savitzky-Golay digital filter to filter the voltage, current, and temperature data in each original charging segment. This filter uses polynomial fitting to fit adjacent data points, smoothing noise while preserving data trend characteristics. Second, the cumulative charging capacity is calculated. Based on the filtered current data and timestamps, the trapezoidal integral method is used to calculate the cumulative charging capacity at each time sampling point from the charging start point, with the initial integral value set to 0. Finally, the filtered voltage, current, and temperature data are combined with the calculated cumulative capacity data to form a filtered charging segment, which is added to the filtered charging segment set. The rationality of the capacity data is verified, and if abnormal jumps occur, the calculation is recalculated.

[0039] Specifically, step S23 involves standardizing the operating conditions of each filtered charging segment in the filtered charging segment set to obtain a normalized charging segment set. It should be understood that due to differences in charging ambient temperature and charging current rate among different filtered charging segments, voltage data at the same SOC are not comparable. For example, voltage may be lower at low temperatures or higher at high currents. Directly using these data for virtual ICA curve reconstruction would result in inconsistent curve shapes, failing to accurately reflect the true aging state of the battery. Therefore, this application further corrects all filtered charging segments to a unified reference operating condition to eliminate the impact of operating condition differences on voltage data. This ensures that voltage data from different charging segments remain consistent under the same operating condition benchmark, guaranteeing the accuracy of subsequent virtual ICA curve reconstruction and the comparability of data between different segments, providing a unified data foundation for extracting characteristic parameters that truly reflect the battery's health state.

[0040] Specifically, in one possible embodiment, step S23 is implemented as follows: First, reference operating condition parameters are set, with the reference temperature set to the standard ambient temperature and the reference charging current rate set to a fixed C-rate. Second, temperature standardization is performed. Based on a pre-calibrated battery equivalent internal resistance-temperature relationship model, the difference in internal resistance between the current segment temperature and the reference temperature is calculated. The filtered voltage data is then corrected to the equivalent voltage at the reference temperature using a voltage compensation formula. Next, current rate standardization is performed. Based on the battery polarization characteristic model, the difference in polarization voltage between the current charging current and the reference current is calculated, further correcting the temperature-standardized voltage data. Finally, the standardized voltage data is combined with the reference temperature, reference current, and cumulative capacity data to form a normalized charging segment and added to the normalized charging segment set.

[0041] In the aforementioned multi-dimensional diagnostic method for the health status of lithium batteries, step S3 involves reconstructing virtual ICA curves and generating confidence scores for the normalized charging segment set to obtain virtual ICA curves and confidence score maps. It should be understood that since each segment in the normalized charging segment set only covers a portion of the voltage range, using any single segment alone cannot obtain a complete incremental capacity analysis (ICA) curve, and the reliability of data from different segments varies. Direct analysis can lead to incomplete or distorted health feature extraction. Therefore, this application reconstructs virtual ICA curves for the normalized charging segments and simultaneously calculates the confidence score for each voltage point to obtain complete and reliability-quantified basic data for ICA analysis. This effectively integrates the valid information from each segment, fills in the missing parts of the voltage range, and quantifies data reliability through confidence scores, providing a basis for subsequent weighted extraction of health features based on confidence scores, and avoiding diagnostic biases caused by incomplete or unreliable data.

[0042] In particular, in one specific embodiment, Figure 4 This is a flowchart of sub-step S3 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application. Figure 4 As shown, step S3 includes: S31, performing ICA data derivation and voltage axis discretization alignment on the normalized charging segment set to obtain a discretized ICA segment set; S32, performing multi-segment information fusion on the discretized ICA segment set to obtain the virtual ICA curve; S33, performing confidence spectrum calculation on the discretized ICA segment set to obtain a confidence spectrum.

[0043] Specifically, step S31 involves performing ICA data derivation and voltage axis discretization alignment on the normalized charging segment set to obtain a discretized ICA segment set. It should be understood that since the normalized charging segment set only contains standardized basic data such as voltage and cumulative capacity, and has not yet been transformed into ICA features reflecting the battery's electrochemical characteristics, and the voltage sampling ranges and densities of different segments differ, direct fusion would lead to data conflicts due to inconsistent benchmarks. Therefore, this application further derives ICA data from the normalized segments and aligns the ICA data of all segments to a unified voltage grid to obtain a discretized ICA segment set with a unified format. This transforms the basic data into electrochemical features directly related to battery aging, eliminates voltage benchmark differences between segments, provides standardized data for subsequent multi-segment information fusion and confidence calculation, and ensures the consistency and integrability of subsequent processing.

[0044] In particular, in one specific embodiment, Figure 5 This is a flowchart of sub-step S31 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application. Figure 5As shown, step S31 includes: S311, performing a validity check on each normalized charging segment in the normalized charging segment set to obtain a validity check result; S312, in response to a passing validity check result, calculating the incremental capacity and voltage midpoint of the normalized charging segment to obtain an ICA point cloud set; S313, performing ICA data grid alignment and vectorization on the ICA point cloud set to obtain a discretized ICA segment set.

[0045] More specifically, step S311 involves performing a validity check on each normalized charging segment in the normalized charging segment set to obtain the validity check result. It should be understood that abnormal data may exist during the acquisition or preprocessing of normalized charging segments, such as voltage jumps, capacity calculation deviations, and temperatures exceeding reasonable ranges. Therefore, this application further verifies the rationality of the voltage, current, capacity, and temperature data of each normalized charging segment to screen out valid segments that meet the analysis requirements. This allows for the early removal of segments containing abnormal data, avoiding interference from abnormal data in subsequent ICA analysis, and ensuring that the data used to derive ICA features are reliable data reflecting the actual charging state of the battery, laying a data quality foundation for the subsequent accurate generation of ICA point clouds.

[0046] Specifically, in one possible embodiment, step S311 is implemented as follows: First, validity check indicators and thresholds are set, including that the voltage must be within the battery's rated operating voltage range, the change in capacity between adjacent sampling points must be less than a specific value, the temperature must be within the battery's normal operating temperature range, and the current fluctuation amplitude must be less than a specific value. Second, each segment in the normalized charging segment set is traversed, and the data is checked point by point to see if it meets the threshold requirements, while simultaneously checking whether the cumulative capacity change trend is continuous without reverse fluctuations. Finally, it is determined whether each segment passes the check, and a list of passed and failed segments and the corresponding validity check results are output. Failed segments are removed from the set.

[0047] More specifically, in step S312, in response to a pass on the validity check result, the incremental capacity and voltage midpoint of the normalized charging segment are calculated to obtain an ICA point cloud set. It should be understood that since the normalized charging segment that passes the validity check only contains basic parameters such as voltage and cumulative capacity, it has not yet been transformed into characteristic data reflecting the internal electrochemical reactions of the battery. Incremental capacity (dQ / dV) and voltage midpoint are core parameters for ICA analysis; without these parameters, subsequent battery health feature extraction cannot be performed. Therefore, this application further calculates the incremental capacity and corresponding voltage midpoint based on the segment data that passes the check, thereby forming a point cloud set containing the core features of ICA. In a specific example of this application, step S312 includes: calculating the incremental capacity and voltage midpoint of the normalized charging segment using the following formula:

[0048]

[0049]

[0050] in, and The first The time sampling point and the first Voltage values ​​at each time sampling point, normalized by temperature and current rate. and For the first The time sampling point and the first The cumulative charging capacity at each time sampling point For the first Incremental capacity corresponding to each time sampling point For the first The midpoint of the voltage corresponding to each time sampling point. In this way, the basic voltage-capacity data can be transformed into electrochemical characteristic data directly related to the battery aging state. Each ICA point cloud can reflect the electrochemical change law during the charging process of the corresponding segment, providing original characteristic data support for subsequent ICA data grid alignment, and ensuring the reliability and effectiveness of the data source for subsequent discretization processing.

[0051] More specifically, step S313 involves aligning and vectorizing the ICA point cloud set using the ICA data grid to obtain a set of discretized ICA segments. It should be understood that since each ICA point cloud is generated based on a single normalized charging segment, the distribution range and density of its voltage midpoints vary. Direct fusion would result in data incompatibility due to inconsistent voltage references, and the point cloud format would be inconvenient for subsequent model input. Therefore, this application further maps the ICA point cloud set to a preset global voltage grid and transforms it into standardized vectors to obtain a set of discretized ICA segments. This allows the ICA features of different segments to be aligned under the same voltage reference, eliminating distribution differences, and simultaneously transforming unstructured point clouds into structured vectors, facilitating information fusion for subsequent input into the model and providing a standardized input format for virtual ICA curve reconstruction.

[0052] Specifically, in one possible embodiment, step S313 is implemented as follows: First, a preset global voltage grid is invoked, which covers the voltage midpoint range of all ICA point clouds with a fixed step size. Second, for each ICA point cloud, an empty vector of the same length as the global voltage grid is created. The "voltage midpoint-incremental capacity" data pairs in the point cloud are traversed, and the index closest to the voltage midpoint in the global grid is found. The incremental capacity value is assigned to this index position. When multiple data pairs are mapped to the same index, the average value is taken. Finally, missing markers are filled in the positions in the vector that are not mapped to data to ensure that the vector dimension is consistent with the grid. All processed vectors are then organized to form a set of discretized ICA fragments.

[0053] Specifically, step S32 involves fusing multi-segment information from the discretized ICA segment set to obtain the virtual ICA curve. It should be understood that since each segment in the discretized ICA segment set only has valid data in certain voltage ranges, with the remaining ranges being missing, using any single segment alone cannot generate a complete ICA curve covering the entire operating voltage range of the battery. Therefore, this application further fuses multi-segment ICA segment information to obtain a complete, smooth virtual ICA curve that conforms to the battery's electrochemical characteristics. This fully integrates the valid information from each segment in different voltage ranges, fills in the missing range data through battery aging patterns learned from the model, and suppresses noise interference from a single segment, generating a more reliable complete ICA curve than any single segment. This provides a high-quality curve foundation for the subsequent accurate extraction of health characteristics such as voltage peak position and peak height.

[0054] In particular, in one specific embodiment, Figure 6 This is a flowchart of sub-step S32 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application. Figure 6 As shown, step S32 includes: S321, inputting each discretized ICA segment in the discretized ICA segment set into the encoder of the pre-trained conditional variational autoencoder to obtain a set of latent vectors; S322, performing a weighted average on the set of latent vectors to obtain a global latent vector; S323, inputting the global latent vector into the decoder of the pre-trained conditional variational autoencoder to obtain a virtual ICA curve.

[0055] More specifically, in step S321, each discretized ICA segment in the set of discretized ICA segments is input into the encoder of a pre-trained conditional variational autoencoder to obtain a set of latent vectors. It should be understood that since discretized ICA segments are high-dimensional vectors and contain a large number of missing data markers, directly fusing multiple segments from high-dimensional vectors faces problems of high computational complexity and significant interference from missing data, making it difficult to effectively extract common ICA information from multiple segments. Therefore, this application further inputs each discretized ICA segment into the encoder of a pre-trained conditional variational autoencoder, transforming it into low-dimensional latent vectors through the encoder's feature compression capability, thereby obtaining a set of latent vectors. This transforms high-dimensional segments with missing data into low-dimensional structured latent vectors, reducing computational complexity. Simultaneously, the encoder's pre-trained parameters filter noise and invalid information, preserving the core ICA features of each segment, providing a concise and effective data form for subsequent weighted fusion of latent vectors.

[0056] Specifically, in one possible embodiment, step S321 is implemented as follows: The encoder, as part of a conditional variational autoencoder, is trained on a dataset containing a large number of high-quality, complete ICA curves collected in the laboratory. This training data covers the entire aging process of the battery from new to end-of-life, and each ICA curve is labeled with its corresponding health state as conditional information. The encoder's network structure can consist of several fully connected layers; for example, the input layer dimension is consistent with the voltage grid dimension, passing through two hidden layers containing ReLU activation functions, and finally outputting to a low-dimensional latent space. Specifically, first, the pre-trained encoder model parameters are loaded. Then, each segment vector in the discretized ICA segment set is traversed, and missing data markers (such as -1) are replaced with zero values ​​to meet the input requirements of the neural network. Finally, the processed segment vector is input into the encoder, and through its nonlinear transformation, a latent vector that highly summarizes the core electrochemical information of the segment is obtained. The latent vectors generated from all segments together form a latent vector set for the next step of weighted fusion.

[0057] More specifically, step S322 involves performing a weighted average of the latent vector set to obtain the global latent vector. It should be understood that since each latent vector in the latent vector set corresponds to a discretized ICA segment, and the effective data volume of different segments varies—some segments have a wide effective voltage range and dense data points, while others have a narrow effective range and sparse data points—a simple arithmetic average would result in segments with less effective information contributing equally to segments with more effective information, reducing the reliability of the fused information. Therefore, this application further calculates weights based on the number of effective data points in each discretized ICA segment and performs a weighted average of the latent vector set to obtain the global latent vector. This allows segments with more effective data and higher information reliability to occupy higher weights during the fusion process, reducing the interference of segments with less effective data on the global information and ensuring that the global latent vector accurately reflects the common ICA features of multiple segments.

[0058] More specifically, step S323 involves inputting the global latent vector into the decoder of a pre-trained conditional variational autoencoder to obtain a virtual ICA curve. It should be understood that since the global latent vector is a low-dimensional structured feature representation, it can only reflect the common features of multiple ICA information segments and cannot be directly used for health feature extraction. Furthermore, the latent vector lacks intuitive electrochemical meaning and needs to be transformed into a high-dimensional ICA curve form. Therefore, this application further inputs the global latent vector into the decoder of a pre-trained conditional variational autoencoder, using the decoder's reconstruction capability to transform it into a high-dimensional ICA curve, thereby obtaining a virtual ICA curve. This allows the low-dimensional latent vector to be restored to an ICA curve with clear electrochemical meaning, ensuring that the curve shape conforms to the ICA characteristics during the actual aging process of the battery, providing a directly analyzable curve carrier for subsequent health feature extraction steps such as peak detection and boundary localization.

[0059] Specifically, in one possible embodiment, step S323 is implemented as follows: First, the decoder module of the pre-trained conditional variational autoencoder is loaded. This decoder and the encoder described in step S321 are trained end-to-end as a single conditional variational autoencoder model, rather than independently. Their training objective is to accurately reconstruct the original complete ICA curve from the latent vectors. The input dimension of this decoder is consistent with the dimension of the global latent vectors, while the output dimension perfectly matches the dimension of the global voltage grid. Second, the weighted and fused global latent vectors are input into the decoder. The decoder utilizes the mapping relationship learned through pre-training to perform feature decoding and dimensionality enhancement on the low-dimensional latent vectors, gradually restoring the complete shape of the high-dimensional ICA curve. Finally, the decoder outputs an incremental capacity vector corresponding to the global voltage grid, i.e., a virtual ICA curve.

[0060] Specifically, step S33 involves calculating a confidence profile for the set of discretized ICA segments to obtain a confidence profile. It should be understood that since the generation of the virtual ICA curve integrates multiple segments of valid data and model-inferred data, the voltage ranges inferred by the model lack real data support, and their reliability cannot be directly judged. If the confidence level of each voltage point is not quantified, unreliable inferred data will be treated the same as reliable real data during subsequent health feature extraction, leading to diagnostic bias. Therefore, this application quantifies the reliability of each voltage point in the virtual ICA curve by statistically analyzing the distribution of valid data in the discretized ICA segments, calculating the confidence level of each voltage point, and generating a confidence profile. This allows for the identification of high-confidence (based on multiple segments of real data) and low-confidence (based on model inference) voltage ranges in the virtual ICA curve, providing a basis for adaptive weighting during subsequent health feature extraction, prioritizing the use of high-confidence data, and reducing the interference of low-confidence data on the diagnostic results.

[0061] In particular, in one specific embodiment, Figure 7 This is a flowchart of sub-step S33 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application. Figure 7 As shown, step S33 includes: S331, performing data density statistics on the discretized ICA fragment set to obtain a discrete count histogram; S332, performing Gaussian smoothing on the discrete count histogram to obtain the confidence spectrum.

[0062] More specifically, step S331 involves performing data density statistics on the discretized ICA segment set to obtain a discrete count histogram. It should be understood that since each segment in the discretized ICA segment set only contains valid data within a certain voltage range, the number of times different voltage points are covered by each segment varies. This coverage frequency directly reflects the reliability basis of the data at that voltage point; without statistical analysis, the true data support level for each voltage point cannot be quantified. Therefore, this application further statistically analyzes the effective data coverage frequency of each voltage point in the discretized ICA segment set to obtain a discrete count histogram reflecting the data support strength of each voltage point. This visually presents the frequency of each voltage point being covered by real data, providing raw data for subsequent confidence level calculations. Voltage points with more coverage frequency correspond to higher potential confidence, while voltage points with fewer or zero coverage frequency correspond to lower potential confidence, ensuring that the subsequent generation of confidence level maps has clear data support.

[0063] Specifically, in one possible embodiment, step S331 is implemented as follows: First, a set of discretized ICA segments aligned to a unified global voltage grid is invoked to define the starting voltage, ending voltage, and step size of the global voltage grid, ensuring that the voltage dimension of all segments is consistent. Second, a zero-based vector of the same length as the global voltage grid is initialized as a counting carrier, where each element of the vector corresponds to a voltage point in the grid. Finally, each segment in the set of discretized ICA segments is traversed, and for each voltage position identified by non-invalid data within the segment, the corresponding element in the counting carrier is incremented by 1. After traversal, the value of each element in the counting carrier represents the number of times the corresponding voltage point is covered by valid data. This counting carrier is a discrete counting histogram. Subsequently, the histogram is checked for integrity to ensure there are no counting anomalies before being used for the next step of processing.

[0064] More specifically, step S332 involves Gaussian smoothing the discrete count histogram to obtain the confidence level spectrum. It should be understood that since the discrete count histogram is a direct statistical result of the number of times effective data covers each voltage point, its values ​​exhibit discrete jump characteristics, meaning that the counts of adjacent voltage points may differ significantly. Directly using this as the confidence level can lead to local fluctuations in the confidence assessment, failing to reflect the continuous change in confidence within the voltage range and affecting the rationality of subsequent weighting of health characteristics. Therefore, this application further employs a Gaussian kernel to perform convolution operations on the discrete count histogram to obtain a smooth confidence level spectrum that reflects the continuous change in confidence. This eliminates the local jumps in discrete counts, making the confidence level spectrum show a continuous and smooth trend on the voltage axis, conforming to the continuous law of battery electrochemical characteristics changing with voltage. This avoids misjudgments of the confidence level in the voltage range due to local count fluctuations. Furthermore, subsequent normalization mapping maps the values ​​to the 0-1 interval, making the confidence level results more intuitive and easier to use for subsequent weighted calculations.

[0065] Specifically, in one possible embodiment, step S332 is implemented as follows: First, the Gaussian kernel parameters required for Gaussian smoothing are determined. A one-dimensional Gaussian kernel with a standard deviation of 5 voltage grid steps is selected. This parameter has been verified through extensive experiments and can achieve effective smoothing while preserving the data density trend. Second, the discrete counting histogram is convolved with the preset Gaussian kernel. Each element of the histogram is traversed, and with that element as the center, adjacent elements are weighted and summed according to the Gaussian kernel weights to obtain the smoothed density value. Finally, the smoothed density map after convolution is subjected to max-min normalization processing. The calculation method is (smoothed density value - minimum density value) / (maximum density value - minimum density value), mapping all values ​​to the 0-1 interval to obtain the final confidence map, ensuring that each value in the map accurately represents the confidence of the corresponding voltage point.

[0066] In the aforementioned multi-dimensional diagnostic method for the health status of lithium batteries, step S4 involves extracting health features from the virtual ICA curve and confidence level spectrum based on confidence level weighting to obtain a weighted health index vector. It should be understood that since the virtual ICA curve contains characteristic peaks reflecting the battery's health status, but the confidence levels of different voltage ranges containing these peaks vary, directly extracting features without considering confidence levels would treat features in low-confidence ranges and high-confidence ranges equally, leading to health indicators that fail to reflect data reliability and affecting subsequent diagnostic accuracy. Therefore, this application further locates and quantifies feature points from the virtual ICA curve, then combines them with the confidence level spectrum to perform weighted fusion of the features, thereby obtaining a weighted health index vector. This ensures that the extracted health indicators not only include the core electrochemical features related to battery aging, but also highlight the contribution of high-confidence features through confidence level weighting, reducing the interference of low-confidence features, providing accurate and reliable indicator input for subsequent multi-dimensional health diagnosis, and significantly improving the accuracy and robustness of the diagnostic results.

[0067] In particular, in one specific embodiment, Figure 8 This is a flowchart of sub-step S4 of the multi-dimensional diagnostic method for the health status of a lithium battery according to an embodiment of this application. Figure 8 As shown, step S4 includes: S41, performing peak detection and boundary localization on the virtual ICA curve to obtain a set of feature points; S42, performing multi-dimensional feature quantization calculation on the virtual ICA curve based on the set of feature points to obtain an original set of health indicators; S43, performing confidence fusion on the original set of health indicators based on the confidence map and the set of feature points to obtain a weighted health indicator vector.

[0068] Specifically, step S41 involves peak detection and boundary localization of the virtual ICA curve to obtain a set of feature points. It should be understood that since the virtual ICA curve is a continuous dQ / dV voltage curve covering the entire voltage range, its core information reflecting the battery's health status is concentrated on the characteristic peaks. However, the curve itself does not explicitly mark the positions and ranges of the peaks; without localization, feature parameters cannot be accurately extracted. Therefore, this application further employs a peak detection algorithm to identify the characteristic peaks in the virtual ICA curve and locate the peak apex and left and right boundaries of each peak to obtain a set of feature points. This transforms key health features in the continuous curve into quantifiable point information, providing a precise spatial range basis for subsequent multi-dimensional feature calculations, ensuring that the parameter extraction of each characteristic peak is limited to an accurate voltage range, and avoiding feature quantification deviations caused by ambiguous peak range definitions.

[0069] Specifically, in one possible embodiment, step S41 is implemented as follows: First, peak detection parameters are set, including peak prominence (to filter out small fluctuations in noise), minimum peak spacing (to avoid misjudging adjacent small peaks), and baseline threshold (to distinguish peaks from background). Second, the voltage sequence of the virtual ICA curve is traversed, and by comparing the changes in dQ / dV values ​​of adjacent points, the local maximum point that meets the parameter requirements is identified as the peak, and its index is recorded. Finally, for each peak, the process is traversed in both the decreasing and increasing voltage directions to find the first point where the dQ / dV value is lower than the baseline threshold as the valley (boundary), and the left and right valley indices are recorded. The left valley, peak, and right valley are combined into a triplet, and all triplets are arranged in ascending voltage order to form a set of feature points, and the rationality of the points is verified.

[0070] Specifically, in step S42, based on the feature point set, multi-dimensional feature quantification calculations are performed on the virtual ICA curve to obtain the original health indicator set. It should be understood that since the feature point set only contains index information of characteristic peaks and does not involve the specific quantification parameters of the peaks, and these parameters are direct indicators reflecting the degree of battery aging, without quantification, the point information cannot be transformed into health indicators that can be used for diagnosis. Therefore, this application further extracts and quantifies the multi-dimensional parameters of each characteristic peak from the virtual ICA curve according to the voltage range defined by the feature point set, thereby obtaining the original health indicator set. In this way, the abstract point index can be transformed into quantitative indicators with clear electrochemical meaning. Each indicator is directly related to the internal aging mechanism of the battery, providing specific analytical objects for subsequent confidence weighting and multi-dimensional diagnosis, ensuring that the diagnostic process has clear numerical basis.

[0071] Specifically, in one possible embodiment, step S42 is implemented as follows: First, each triplet in the feature point set, namely the left valley index, peak index, and right valley index, is called to extract the dQ / dV value sequence and corresponding voltage sequence within that interval from the virtual ICA curve. Second, features in each dimension are calculated: peak position is the voltage value corresponding to the peak index, peak height is the dQ / dV value corresponding to the peak index, peak area is the integral of the dQ / dV value within the interval with the voltage step size, and half-width at half-maximum (WHM) is the voltage difference corresponding to half the peak height. Finally, the position, height, area, and WHM parameters of each feature peak are encapsulated into a structure. All structures are arranged in ascending order of feature peak voltage to form the original health indicator set, and each indicator is numerically verified to ensure no outliers.

[0072] Specifically, in step S43, based on the confidence level map and the feature point set, the original health indicator set is fused with confidence levels to obtain a weighted health indicator vector. It should be understood that since each indicator in the original health indicator set only reflects the quantitative attributes of the characteristic peak and does not consider the data reliability of its voltage range, if the original indicators are used directly, the indicators in low-reliability ranges will contribute equally to the indicators in high-reliability ranges, leading to the diagnostic results being interfered with by unreliable data. Therefore, this application further determines the voltage range corresponding to each indicator based on the feature point set, extracts the confidence level of that range, and uses it as a weight to fuse with the original indicators to obtain the weighted health indicator vector. This allows the original indicators to be deeply bound to data reliability, with high-reliability indicators having a higher weight in subsequent diagnoses, and the influence of low-reliability indicators being weakened. This ensures that the weighted indicator vector better reflects the true health status of the battery, providing a guarantee for the reliability of multi-dimensional diagnostic results.

[0073] Specifically, in one possible embodiment, step S43 is implemented as follows: First, each indicator in the original health indicator set is traversed, and the voltage range of the characteristic peak is determined based on its corresponding feature point triplet. Second, all confidence values ​​within this range are extracted from the confidence spectrum, and their arithmetic mean is calculated as the confidence weight of the indicator. Third, the position, height, area, and half-width at half-maximum (WHM) parameters of each original indicator are multiplied by their corresponding confidence weights to obtain weighted feature parameters. Finally, the feature parameters are arranged in ascending order of characteristic peak voltage, and all weighted feature parameters are flattened into a one-dimensional vector. The number and value range of vector elements are verified to ensure no missing or outlier values, thus forming a weighted health indicator vector.

[0074] In particular, in a preferred embodiment, step S43 includes: for each original health indicator in the original health indicator set, obtaining the voltage range defined by its corresponding characteristic peak; determining the weight of each voltage point within the voltage range according to the virtual ICA curve, wherein the larger the amplitude corresponding to each voltage point on the virtual ICA curve, the greater the weight of that voltage point; and, based on the determined weights, performing a weighted average of each confidence value corresponding to the voltage range in the confidence spectrum to obtain the weighted confidence of the original health indicator, and using it to generate the weighted health indicator vector.

[0075] Specifically, when performing confidence fusion on the original set of health indicators, if the arithmetic mean is used to calculate the interval confidence, then each voltage point within the boundary of the characteristic peak contributes equally to the feature. However, from the perspective of electrochemistry and signal processing, the diagnostic value of ICA characteristic peaks is mainly concentrated in their most significant parts, namely the peak top and its surrounding area, because these areas represent the voltage plateaus where the electrochemical reaction is most intense and the features are most obvious. Although the shoulders and valleys of the peaks define the range of the features, their dQ / dV values ​​are small, and their contribution to the core attributes of the features (such as peak height and peak position) is relatively small.

[0076] Therefore, if an arithmetic mean is used, assigning the same weight to the confidence levels of regions with high dQ / dV values ​​at the peaks and those with low dQ / dV values ​​at the valleys can lead to an artificially inflated and misleading overall confidence score if the model is highly uncertain about the generation of peak regions (low confidence) but has high confidence in broad, flat baseline regions. Furthermore, the arithmetic mean cannot utilize known morphological information; for example, the internal information distribution of a tall, narrow peak differs from that of a short, wide peak, and the confidence score calculation should reflect this morphological difference.

[0077] Based on this, a weighted average mechanism using the amplitude of the ICA curve itself as a weight is introduced to calculate the interval confidence level. That is, within the voltage range of the characteristic peak, the larger the dQ / dV value of a point, the greater its contribution to the formation of the characteristic peak, and its corresponding confidence level should have a higher weight in the final comprehensive confidence assessment. This shifts the focus of confidence assessment from the average confidence level across the entire voltage range to the confidence level in the core morphological part constituting the characteristic peak, thus naturally coupling confidence assessment with the extracted features, since both features are dominated by points with high dQ / dV values.

[0078] First, for each original health indicator in the original health indicator set, the voltage range defined by its corresponding characteristic peak is obtained. Specifically, the original health indicator set is processed according to the voltage order of peak appearance. and feature site set Sort the raw health indicators and iterate through each of the sorted raw indicators. and their corresponding characteristic sites From the virtual ICA curve Extract from arrive The defined dQ / dV subsequence, for example denoted as Furthermore, to ensure the weights are positive, the subsequence is baseline-shifted to make all its values ​​non-negative. Typically, the ICA value itself is non-negative, and this step serves as a robustness measure.

[0079] Then, based on the virtual ICA curve, the weight of each voltage point within the voltage range is determined. The larger the amplitude corresponding to each voltage point on the virtual ICA curve, the greater the weight of that voltage point. That is, the calculation... Each point in weight Weight It is proportional to the dQ / dV value at that point, and by summing and normalizing, it ensures that the sum of all weights is 1.

[0080]

[0081] in, and These are the first virtual ICA curves. The and the first The dQ / dV values ​​at each point and These are the right boundary index and left boundary index of the feature site, respectively. For the index variable in the summation process, for The Middle The weighting of each point ensures that high dQ / dV points, which are more critical for health diagnosis, have a higher weight in subsequent confidence assessments, preventing low-contribution marginal points from diluting core information and significantly improving the relevance and accuracy of interval confidence assessments.

[0082] Then, based on the determined weights, a weighted average is calculated for each confidence value corresponding to the voltage interval in the confidence spectrum to obtain the weighted confidence level of the original health indicator:

[0083]

[0084] in, The first in the confidence graph Confidence values ​​for each point For the index variable in the summation process, For the first The weighted confidence levels correspond to the original health indicators. This ensures that the confidence levels of core high dQ / dV points dominate the interval confidence levels, preventing situations where the core area is unreliable but the overall confidence level is artificially high, and ensuring that the interval confidence levels accurately reflect the credibility level of key areas.

[0085] Finally, The four features and the newly calculated weighted confidence level These features are combined into a quintuple to obtain a weighted health index vector after traversing all feature peaks. This allows the subsequent diagnostic model to prioritize high-confidence features during analysis, significantly reducing the interference of low-confidence features on diagnostic results and greatly improving the accuracy and robustness of multi-dimensional health diagnosis.

[0086] Therefore, the improved confidence level It will be more sensitive to the confidence profile values ​​of the peak region. That is, if the generative model's prediction of the peak lacks real data support (i.e., the peak confidence is low), even if the confidence of the base portion of the peak is high, the final result will be uncertain. It will also be significantly lowered, thereby increasing the sensitivity of key areas and more accurately reflecting the uncertainty of key parts of the feature.

[0087] In addition, if the generative model produces some unrealistic broadening or tailing in the low confidence region, the dQ / dV values ​​in these regions are usually low. Under the weighted average, the low confidence values ​​in these low dQ / dV regions also have a small weight, thereby reducing their negative impact on the final confidence score. This makes the evaluation results focus more on the main part of the feature and enhances the robustness to feature morphological distortion.

[0088] In the aforementioned multi-dimensional diagnostic method for the health status of lithium batteries, step S5 involves performing a multi-dimensional health status diagnosis on the weighted health indicator vector to obtain a multi-dimensional SOH report. It should be understood that although the weighted health indicator vector integrates confidence levels and feature parameters, it is only in a structured numerical form and has not been transformed into a health assessment conclusion with electrochemical significance and engineering value. Furthermore, it cannot directly present key health information such as battery capacity decay, internal resistance changes, and aging mechanisms, making it difficult to meet users' needs for a comprehensive understanding of battery health status and decision-making guidance. Therefore, this application further relies on a diagnostic model and expert rule base to conduct multi-dimensional analysis of the weighted health indicator vector to generate a multi-dimensional SOH report that includes core SOH parameters, aging mechanism attribution, and safety risk warnings. This transforms the abstract feature vector into a concrete and systematic health assessment result, providing both traditional quantitative values ​​for capacity SOH and internal resistance SOH, and tracing the dominant factors of aging and warning of potential safety risks. This provides a direct basis for battery maintenance planning, remaining life prediction, and safety management, significantly improving the engineering practicality and decision support value of the diagnostic results.

[0089] Specifically, in one possible embodiment, step S5 is implemented as follows: First, a pre-trained multi-dimensional diagnostic model and expert rule base are loaded. The model includes a gradient boosting tree regression sub-model for SOH estimation and a support vector machine classification sub-model for aging mechanism attribution. The rule base has built-in mapping logic between feature peak parameters and aging modes. Second, a weighted health index vector is input into the regression sub-model. The model combines the confidence weights of each feature in the vector to output specific values ​​for capacity SOH and internal resistance SOH. Then, the vector is input into the classification sub-model and matched with the expert rule base. Based on parameters such as feature peak position offset and peak height attenuation rate, the current dominant aging mechanism is determined and its contribution to capacity attenuation is quantified. Finally, combining the high-voltage range feature peak parameters and corresponding confidence levels in the vector, the safety risk levels such as lithium plating are assessed. All diagnostic results and the credibility of each conclusion are integrated to form a multi-dimensional SOH report. After clarifying the basis and reference standards for each result, the report is output to the designated terminal.

[0090] In summary, a multi-dimensional diagnostic method for the health status of lithium batteries based on embodiments of this application is explained. First, discrete charging data fragments are obtained from the vehicle management system. Based on these fragments, a virtual incremental capacity analysis curve is intelligently reconstructed, and a confidence score is simultaneously generated to quantify the reliability of the reconstructed data. Then, key health features are extracted from the reconstructed curves, and these features are adaptively weighted using the confidence score, thereby generating a comprehensive index vector that accurately reflects the true aging state of the battery. Finally, deep diagnostics are performed based on this index vector to achieve a multi-dimensional and accurate diagnosis of the battery's health status, outputting a multi-dimensional battery health status assessment report. This effectively overcomes the interference of poor data quality under real-world driving conditions on diagnostic accuracy, significantly improving the practical applicability and reliability of the diagnosis.

[0091] Figure 9 This is a block diagram of a multi-dimensional diagnostic system for the health status of a lithium battery according to an embodiment of this application. Figure 9 As shown, the multi-dimensional health status diagnosis system 100 for lithium batteries according to an embodiment of this application includes: a raw BMS data acquisition module 110 for acquiring raw BMS data; a charging data preprocessing module 120 for extracting charging data segments from the raw BMS data and preprocessing them to obtain a normalized charging segment set; an ICA reconstruction and confidence generation module 130 for reconstructing virtual ICA curves and generating confidence scores on the normalized charging segment set to obtain virtual ICA curves and confidence scores; a health feature extraction module 140 for extracting health features based on confidence scores from the virtual ICA curves and confidence scores to obtain a weighted health index vector; and a multi-dimensional health status diagnosis module 150 for performing multi-dimensional health status diagnosis on the weighted health index vector to obtain a multi-dimensional SOH report.

[0092] As described above, the multi-dimensional diagnostic system 100 for lithium battery health status according to embodiments of this application can be implemented in various wireless terminals, such as servers with multi-dimensional diagnostic algorithms for lithium battery health status. In one possible implementation, the multi-dimensional diagnostic system 100 for lithium battery health status according to embodiments of this application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the multi-dimensional diagnostic system 100 for lithium battery health status can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the multi-dimensional diagnostic system 100 for lithium battery health status can also be one of many hardware modules of the wireless terminal.

[0093] Alternatively, in another example, the multi-dimensional diagnostic system 100 for the health status of the lithium battery and the wireless terminal may also be separate devices, and the multi-dimensional diagnostic system 100 for the health status of the lithium battery may be connected to the wireless terminal via wired and / or wireless networks and transmit interactive information in accordance with an agreed data format.

[0094] Here, those skilled in the art will understand that the specific operations of each step in the above-mentioned multi-dimensional diagnostic system for the health status of lithium batteries have been referenced above. Figures 1 to 8 The multi-dimensional diagnostic method for the health status of lithium batteries has been described in detail, and therefore, its repeated description will be omitted.

Claims

1. A multi-dimensional diagnostic method for the health status of lithium batteries, characterized in that, include: Obtain raw BMS data; Charging data segments are extracted from the raw BMS data and preprocessed to obtain a normalized set of charging segments. Virtual ICA curves and confidence levels are reconstructed and generated for the normalized charging segment set to obtain virtual ICA curves and confidence levels. We extract health features based on confidence weighting from the virtual ICA curve and confidence profile to obtain a weighted health index vector; Multidimensional health status diagnosis is performed on the weighted health indicator vector to obtain a multidimensional SOH report.

2. The multi-dimensional diagnostic method for the health status of lithium batteries according to claim 1, characterized in that, Charging data segments are extracted from the raw BMS data and preprocessed to obtain a normalized set of charging segments, including: The original BMS data is used to identify and extract charging segments to obtain the original set of charging segments; Data filtering and cumulative capacity calculation are performed on each original charging segment in the original charging segment set to obtain the filtered charging segment set. The working conditions of each filtered charging segment in the set of filtered charging segments are standardized to obtain a normalized charging segment set.

3. The multi-dimensional diagnostic method for the health status of lithium batteries according to claim 2, characterized in that, Virtual ICA curve reconstruction and confidence generation are performed on the normalized charging segment set to obtain virtual ICA curves and confidence maps, including: The normalized charging segment set is subjected to ICA data derivation and voltage axis discretization alignment to obtain a discretized ICA segment set; The virtual ICA curve is obtained by fusing multi-segment information from the discretized ICA segment set. Confidence maps are calculated on the discretized ICA fragment set to obtain the confidence maps.

4. The multi-dimensional diagnostic method for the health status of lithium batteries according to claim 3, characterized in that, ICA data derivation and voltage axis discretization alignment are performed on the normalized charging segment set to obtain a discretized ICA segment set, including: A validity check is performed on each normalized charging segment in the set of normalized charging segments to obtain the validity check results; In response to a pass validity check, the incremental capacity and voltage midpoint of the normalized charging segment are calculated to obtain the ICA point cloud set; The ICA point cloud set is aligned with the ICA data grid and vectorized to obtain a set of discretized ICA fragments.

5. The multi-dimensional diagnostic method for the health status of lithium batteries according to claim 4, characterized in that, In response to a successful validity check, the incremental capacity and voltage midpoint of the normalized charging segment are calculated to obtain an ICA point cloud set, including: calculating the incremental capacity and voltage midpoint of the normalized charging segment using the following formula: ； ； in, and The first The time sampling point and the first Voltage values ​​at each time sampling point, normalized by temperature and current rate. and For the first The time sampling point and the first The cumulative charging capacity at each time sampling point For the first Incremental capacity corresponding to each time sampling point For the first The midpoint of the voltage corresponding to each time sampling point.

6. The multi-dimensional diagnostic method for the health status of lithium batteries according to claim 3, characterized in that, The virtual ICA curve is obtained by fusing multi-segment information from a set of discretized ICA segments, including: Each discretized ICA segment in the set of discretized ICA segments is input into the encoder of the pre-trained conditional variational autoencoder to obtain the set of latent vectors. A weighted average of the set of latent vectors is used to obtain the global latent vectors; The global latent vector is input into the decoder of the pre-trained conditional variational autoencoder to obtain the virtual ICA curve.

7. The multi-dimensional diagnostic method for the health status of lithium batteries according to claim 3, characterized in that, Confidence maps are calculated on the discretized ICA fragment set to obtain the confidence maps, including: Data density statistics are performed on the discretized ICA fragment set to obtain a discrete count histogram; The discrete count histogram is Gaussian smoothed to obtain the confidence spectrum.

8. The multi-dimensional diagnostic method for the health status of lithium batteries according to claim 1, characterized in that, Confidence-weighted health feature extraction is performed on the virtual ICA curve and confidence profile to obtain a weighted health indicator vector, including: Peak detection and boundary localization are performed on the virtual ICA curve to obtain a set of feature points; Based on the set of feature points, multi-dimensional feature quantification calculations are performed on the virtual ICA curve to obtain the original set of health indicators; Based on the confidence map and feature point set, the original health indicator set is fused with confidence to obtain a weighted health indicator vector.

9. The multi-dimensional diagnostic method for the health status of lithium batteries according to claim 8, characterized in that, Based on the confidence profile and feature point set, the original health indicator set is fused to obtain a weighted health indicator vector, including: For each original health indicator in the set of original health indicators, obtain the voltage range defined by its corresponding characteristic peak. Based on the virtual ICA curve, the weight of each voltage point within the voltage range is determined, wherein the larger the amplitude corresponding to each voltage point on the virtual ICA curve, the greater the weight of that voltage point. Based on the determined weights, the confidence values ​​of each voltage interval in the confidence spectrum are weighted and averaged to obtain the weighted confidence of the original health index, which is then used to generate the weighted health index vector.

10. A multi-dimensional diagnostic system for the health status of a lithium battery, characterized in that, include: The raw BMS data acquisition module is used to acquire raw BMS data; The charging data preprocessing module is used to extract charging data segments from the raw BMS data and preprocess them to obtain a normalized charging segment set. The ICA reconstruction and confidence generation module is used to reconstruct virtual ICA curves and generate confidence scores on a normalized charging segment set to obtain virtual ICA curves and confidence scores. The health feature extraction module is used to extract health features from the virtual ICA curve and confidence map based on confidence weighting to obtain a weighted health index vector; The multi-dimensional health status diagnosis module is used to perform multi-dimensional health status diagnosis on the weighted health indicator vector to obtain a multi-dimensional SOH report.

Images