Battery cell data analysis method and device based on multi-dimensional fusion and storage medium

Abstract

The battery cell data analysis method based on multi-dimensional fusion provided by the embodiment of the invention comprises the following steps: synchronously preprocessing multi-dimensional heterogeneous original data from a battery cell; inputting the preprocessed multi-dimensional data into a cross-modal depth feature fusion network to generate a depth fusion feature vector; inputting the deep fusion feature vector into a pre-trained main evaluation model and a lightweight incremental learning model to obtain a first analysis result and a second analysis result respectively; according to a preset trigger condition and a confidence degree evaluation result of the main evaluation model, whether new data with a truth value label is used for carrying out parameter updating on the lightweight incremental learning model is determined, and parameters of the main evaluation model are kept unchanged in the updating process; and according to the real-time confidence of the main evaluation model, carrying out dynamic weighted fusion on the first analysis result and the second analysis result to obtain a battery cell analysis result. The method aims at improving the estimation precision of key parameters such as the cell health state and the residual life.

Description

Technical Field

[0001] This application belongs to the field of battery management technology, and in particular relates to a method, device and storage medium for cell data analysis based on multi-dimensional fusion. Background Technology

[0002] With the rapid development of electric vehicles and large-scale energy storage technologies, the performance degradation and safety issues of batteries, as the core energy carrier, have attracted much attention. Currently, battery management systems (BMS) collect data such as voltage, current, and temperature and independently analyze single-dimensional data or perform simple early fusion to assess the cell state in real time. However, due to the multi-source (BMS sensors, battery testers), heterogeneous (time-series signals, scalars, images), and multi-scale (second-level current, hour-level capacity) characteristics of cell data, information loss will occur if the deep coupling relationships and cross-scale correlations between different physical quantities (such as voltage and temperature) are not deeply explored, limiting the accuracy of cell state estimation. Therefore, there is an urgent need for a cell state analysis method that can deeply fuse multi-source heterogeneous information and adaptively evolve throughout the entire cell life cycle to improve the accuracy of cell state analysis. Summary of the Invention

[0003] In view of this, embodiments of this application provide a method, device and storage medium for battery cell data analysis based on multi-dimensional fusion. By using a cross-modal attention mechanism to perform deep feature fusion on multi-dimensional data of the battery cell, the inherent correlation between multi-dimensional data is fully explored, providing a rich and discriminative information foundation for battery cell status assessment, and significantly improving the estimation accuracy of key parameters such as battery cell health status and remaining life.

[0004] This application provides a method for analyzing battery cell data based on multi-dimensional fusion, including: synchronously preprocessing multi-dimensional heterogeneous raw data from battery cells; The preprocessed multi-dimensional data is input into a cross-modal deep feature fusion network to generate a deep fusion feature vector; The deep fusion feature vector is input into the pre-trained main evaluation model and the lightweight incremental learning model to obtain the first analysis result and the second analysis result, respectively. Based on the preset triggering conditions and the confidence evaluation results of the main evaluation model, it is determined whether to use new data with truth labels to update the parameters of the lightweight incremental learning model, and the parameters of the main evaluation model remain unchanged during the update process. Based on the real-time confidence level of the main evaluation model, the first analysis result and the second analysis result are dynamically weighted and fused to obtain the cell analysis result.

[0005] In one embodiment, the multi-dimensional heterogeneous raw data includes at least electrical dimension data, thermal dimension data, time dimension data, and operating condition dimension data.

[0006] In one embodiment, the cross-modal deep feature fusion network includes: multiple feature extraction branches adapted to different data dimensions and a cross-modal attention fusion module; the cross-modal attention fusion module is used to dynamically learn and weightedly fuse the features output by each feature extraction branch to generate the deep fusion feature vector.

[0007] In one embodiment, the triggering condition includes at least one of the following: The confidence level of the main evaluation model is lower than the first threshold, and the true value label of the cell state is obtained. Or a significant shift in the feature distribution of the input data is detected.

[0008] In one embodiment, updating the parameters of the lightweight incremental learning model includes: updating the parameters of the lightweight incremental learning model by using elastic weight fixed constraints.

[0009] In one embodiment, the method further includes: The knowledge of the lightweight incremental learning model is periodically distilled into the main evaluation model for incremental updates to the main evaluation model.

[0010] In one embodiment, the cell analysis results are used to indicate the cell's health status, remaining service life, or fault warning information.

[0011] A second aspect of this application provides a battery cell data analysis device based on multi-dimensional fusion, comprising: The data preprocessing module is used to synchronously preprocess multi-dimensional heterogeneous raw data from battery cells; The feature fusion module is used to input preprocessed multi-dimensional data into a cross-modal deep feature fusion network to generate a deep fusion feature vector. The collaborative analysis module is used to input the deep fusion feature vector into the pre-trained main evaluation model and the lightweight incremental learning model to obtain the first analysis result and the second analysis result, respectively. The incremental learning module is used to determine whether to use new data with truth labels to update the parameters of the lightweight incremental learning model based on preset trigger conditions and the confidence evaluation results of the main evaluation model. The parameters of the main evaluation model remain unchanged during the update process. The module is used to dynamically weight and fuse the first analysis result and the second analysis result based on the real-time confidence level of the main evaluation model to obtain the cell analysis result.

[0012] In one embodiment, the multi-dimensional heterogeneous raw data includes at least electrical dimension data, thermal dimension data, time dimension data, and operating condition dimension data.

[0013] In one embodiment, the cross-modal deep feature fusion network includes: multiple feature extraction branches adapted to different data dimensions and a cross-modal attention fusion module; the cross-modal attention fusion module is used to dynamically learn and weightedly fuse the features output by each feature extraction branch to generate the deep fusion feature vector.

[0014] In one embodiment, the triggering condition includes at least one of the following: The confidence level of the main evaluation model is lower than the first threshold, and the true value label of the cell state is obtained. Or a significant shift in the feature distribution of the input data is detected.

[0015] In one embodiment, updating the parameters of the lightweight incremental learning model includes: updating the parameters of the lightweight incremental learning model by using elastic weight fixed constraints.

[0016] In one embodiment, the device further includes: The distillation module is used to periodically distill the knowledge of the lightweight incremental learning model into the main evaluation model for progressive updates of the main evaluation model.

[0017] In one embodiment, the cell analysis results are used to indicate the cell's health status, remaining service life, or fault warning information.

[0018] A third aspect of this application provides a battery cell data analysis device based on multi-dimensional fusion, characterized in that it includes: a processor, a memory, and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, it implements the steps of the method described in the first aspect above.

[0019] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in the first aspect above.

[0020] The multi-dimensional fusion-based battery cell data analysis method provided in this application includes: synchronously preprocessing multi-dimensional heterogeneous raw data from the battery cell; inputting the preprocessed multi-dimensional data into a cross-modal deep feature fusion network to generate a deep fusion feature vector; inputting the deep fusion feature vector into a pre-trained main evaluation model and a lightweight incremental learning model to obtain a first analysis result and a second analysis result, respectively; determining whether to use new data with truth labels to update the parameters of the lightweight incremental learning model based on preset triggering conditions and the confidence evaluation result of the main evaluation model, while keeping the parameters of the main evaluation model unchanged during the update process; and dynamically weighting and fusing the first analysis result and the second analysis result based on the real-time confidence of the main evaluation model to obtain the battery cell analysis result. By using a cross-modal attention mechanism to perform deep feature fusion on the multi-dimensional data of the battery cell, the inherent correlation between the multi-dimensional data is fully explored, providing a rich and discriminative information foundation for battery cell status evaluation, and significantly improving the estimation accuracy of key parameters such as battery cell health status and remaining lifespan. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a multi-dimensional fusion-based cell data analysis method provided in an embodiment of this application; Figure 2 A schematic diagram of the structure of a cell data analysis device based on multi-dimensional fusion provided in an embodiment of this application; Figure 3 This is a schematic diagram of a cell data analysis device based on multi-dimensional fusion, provided as an embodiment of this application. Detailed Implementation

[0023] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0025] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0026] In this document, the term "embodiment" means 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 separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0027] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0028] In the description of the embodiments of this application, the term "multiple frames" refers to two or more (including two).

[0029] In the description of the embodiments of this application, the technical terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of this application and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0030] This application provides a multi-dimensional fusion-based cell data analysis method. By using a cross-modal attention mechanism to perform deep feature fusion on multi-dimensional cell data, it fully explores the intrinsic correlation between multi-dimensional data, providing a rich and discriminative information foundation for cell status assessment, and significantly improving the estimation accuracy of key parameters such as cell health status and remaining life.

[0031] Please see Figure 1 As shown, Figure 1 This is a flowchart illustrating a multi-dimensional fusion-based cell data analysis method provided in an embodiment of this application. Figure 1As can be seen, the cell data analysis method based on multi-dimensional fusion in this application includes steps S110 to S150. Details are as follows: S110: Synchronous preprocessing of multi-dimensional heterogeneous raw data from battery cells.

[0032] The multi-dimensional heterogeneous raw data consists of raw time-series data streams from multiple data sources such as battery management system (BMS), temperature sensor, and operating condition recorder, including at least electrical dimension data, thermal dimension data, time dimension data, and operating condition dimension data.

[0033] Specifically, electrical dimension data includes high-frequency continuous signals, such as individual cell / total voltage (sampling rate fs≥10 Hz) and current (fs≥10 Hz), reflecting the dynamics of electrochemical reactions. Thermal dimension data includes mid-to-low frequency continuous / discrete signals, such as the temperature of multiple temperature measurement points arranged on the cell surface, terminals, and within the module (fs≈0.1-1 Hz), reflecting heat distribution and thermal management status. Time dimension data includes low-frequency discrete scalars, such as cycle count (each cycle +1) and calendar days (daily +1), characterizing the long-term aging process. Operating condition dimension data includes event-driven discrete or continuous signals, such as charge / discharge rate (C-rate), state of charge / discharge (SOC) range, resting time, and ambient temperature, describing the usage scenario.

[0034] For example, the multi-dimensional heterogeneous raw data of the battery cells are preprocessed synchronously, including time synchronization alignment, data cleaning and quality verification, and data slicing and standardization, to obtain structured data.

[0035] The time synchronization alignment process includes: using the timestamps of the electrical dimension data with the highest sampling rate as the reference axis; for sequences with lower frequencies but important morphology, such as thermal sequences, nonlinear alignment is performed with electrical signals (e.g., voltage curves) on the reference axis. By finding the path with the minimum cumulative distance, points of the thermal sequence are mapped onto the electrical time axis, preserving the correspondence between its physical processes (e.g., temperature rise) and electrical processes. It should be noted that this step is not simple interpolation. For time dimension (e.g., cycle number) and operating condition dimension (e.g., multiplier) data, forward padding is used for time alignment on the reference time axis. For example, all time points within a cycle share the same cycle number; the multiplier remains constant between changing moments.

[0036] The data cleaning and quality verification process includes: using a combination of physical constraint-based rules (e.g., voltage must be between [0, upper limit]) and statistical methods (e.g., the 3σ principle) to identify and remove jump points caused by sensor malfunctions or interference. Removed points are replaced by linear interpolation of the preceding and following valid data. Short-term missing data (< a set duration) are handled using linear or spline interpolation; long-term missing data is marked as invalid and not used for subsequent analysis. For high-frequency electrical signals, a sliding window mid-range filter (window length 5-10 points) is used to remove impulse noise, followed by a low-pass Butterworth filter (cutoff frequency set according to the cell's dynamic response, e.g., 1 Hz) to smooth high-frequency noise while preserving the main trends.

[0037] Ultimately, each data point is standardized into a structured object, and all data forms a multidimensional data matrix. For example, the structured object includes a synchronization timeline (e.g., an equally spaced time series of length N), a multidimensional data matrix (e.g., a matrix of shape [N, D], where D is the total number of features (e.g., voltage + current + 6 temperatures + ...)), and sample metadata (e.g., record cycle count, slice type, and data quality identifier). Slice types include fast charging, slow discharging, etc.

[0038] S120: Input the preprocessed multi-dimensional data into the cross-modal deep feature fusion network to generate a deep fusion feature vector.

[0039] The cross-modal deep feature fusion network includes: multiple feature extraction branches adapted to different data dimensions and a cross-modal attention fusion module; wherein, the cross-modal attention fusion module is used to dynamically learn and weightedly fuse the features output by each feature extraction branch to generate a deep fusion feature vector.

[0040] Specifically, the feature extraction branches include: an electrical feature extraction branch, used to adapt to a two-dimensional time-series signal formed by splicing voltage and current sequences. This branch uses a one-dimensional convolutional layer to extract local charge-discharge curve morphological features (such as voltage plateaus and differential voltage dV / dt), followed by a gated recurrent unit (GRU) to capture long-range charge-discharge dynamic dependencies, ultimately outputting an electrical feature vector. A thermal feature extraction branch is used to adapt to multi-point temperature monitoring data of the battery cell. In a preferred embodiment, each temperature measurement point and its spatial relationship are constructed as graph-structured data, where nodes are temperature measurement points, node features are temperature values, and edges are defined by physical location distance or a heat conduction model. This branch uses a graph convolutional network (GCN) for two-layer message propagation to aggregate neighborhood temperature information and output a feature vector characterizing heat distribution uniformity, hotspot locations, and heat diffusion rates. In another preferred embodiment, for temperature data with unclear spatial relationships, this branch uses a fully connected network to process the temperature vector to extract features. The spatiotemporal and operating condition feature extraction branch is used to adapt scalar or discrete data such as cycle number (scalar), average charge / discharge rate (scalar), and state of charge interval (using one-hot encoding). This branch first maps discrete operating condition parameters into continuous vectors through an embedding layer, then concatenates them with the time scalar, and finally extracts high-level operating condition context feature vectors through a fully connected layer.

[0041] The cross-modal attention fusion module is used to dynamically weight and fuse the heterogeneous feature vectors output from the above branches. Its workflow includes the following three steps: A, B, and C. A. Projection: The feature vectors input from each branch are projected onto a unified semantic space through independent linear layers to obtain projected feature vectors { }

[0042] B. Attention Weight Calculation: Introduce a learnable query vector Q associated with a specific analytical task (e.g., health status estimation). Projected feature vectors for each modality. Calculate its compatibility score with Q, and obtain the attention weights after Softmax normalization. The specific calculation formula is as follows: Where Wi is a learnable linear transformation matrix (used to transform...). (Converted to a key vector), where d is the feature dimension and √d is the scaling factor. Attention weights. The dynamics reflect the importance of the i-th modal feature in the current task context and the specific state of the battery cell.

[0043] C. Weighted Fusion and Output: Based on the calculated attention weights, the projected feature vectors are weighted and summed. Subsequently, the fused vector The input is fed into a fully connected layer containing an activation function, where it undergoes nonlinear transformation and dimensionality reduction, ultimately outputting a fixed-length deep fusion feature vector (e.g., 128-dimensional or 256-dimensional).

[0044] By adopting the above structure, the network can adaptively focus on the data modalities most relevant to the current analysis objective, thereby achieving deep information fusion.

[0045] S130: Input the deep fusion feature vector into the pre-trained main evaluation model and the lightweight incremental learning model to obtain the first analysis result and the second analysis result, respectively.

[0046] The deep fusion feature vector generated in step S120 is input in parallel into the pre-trained main evaluation model and the lightweight incremental learning model to generate the first analysis result and the second analysis result, respectively.

[0047] The pre-trained master evaluation model provides benchmark analysis capabilities based on the aging patterns of battery cells. This master evaluation model is a deep neural network model, and its network parameters are optimized and fixed through offline supervised training on a large-scale historical battery cell aging dataset. This historical dataset covers the complete lifecycle data of the same model or type of battery cells under various operating conditions, enabling the model to learn and internalize common patterns of battery cell performance degradation.

[0048] In a preferred embodiment, the main evaluation model can be a multilayer perceptron or a Transformer encoder with a self-attention mechanism, which has sufficient parameter capacity to characterize the complex nonlinear mapping relationship between cell state and multi-dimensional fused features.

[0049] The main evaluation model takes the deep fusion feature vector as input and outputs the first analysis result. This first analysis result can specifically be represented as an estimate of the cell's health status, a predicted remaining service life, or a warning probability of various faults.

[0050] It should be noted that during online operation, the network parameters of the master evaluation model remain unchanged (i.e., in a frozen state). This serves as a stable reference benchmark, providing an initial judgment with good generalization for cell condition analysis.

[0051] The lightweight incremental learning model is used to learn online and adapt to the unique aging trajectory and performance deviation of each monitored cell. This lightweight incremental learning model is a machine learning model with a simplified structure and significantly fewer parameters than the main evaluation model, enabling rapid training and low-computational-cost online updates. Its network parameters can be initialized with small random values ​​or with simplified parameters obtained from knowledge distillation of the main evaluation model.

[0052] In a preferred embodiment, the lightweight incremental learning model can be a stochastic vector functional link network, a linear regression model, or a shallow neural network with very few layers. The input to this lightweight incremental learning model is also the deep fusion feature vector, and the output is a second analysis result. The representation of the second analysis result corresponds to the first analysis result. During online operation, the network parameters of the lightweight incremental learning model can be dynamically updated according to the rules of subsequent steps (S140). Its core function is to quickly capture and fit the individualized deviations exhibited by the current battery cell from the group average pattern on which the main evaluation model is based, using a small amount of newly generated data with truth labels.

[0053] Through the aforementioned dual-model collaborative architecture, this step provides preliminary results for cell state analysis that combine global prior knowledge stability with local individual adaptability, laying the foundation for subsequent intelligent fusion and decision-making.

[0054] S140: Based on the preset triggering conditions and the confidence evaluation results of the main evaluation model, determine whether to use new data with truth labels to update the parameters of the lightweight incremental learning model. The parameters of the main evaluation model remain unchanged during the update process.

[0055] This step aims to achieve safe and conservative online optimization of the lightweight incremental learning model, ensuring its adaptability to individual cell variations while maximizing the retention of learned knowledge and avoiding catastrophic forgetting. The parameters of the main evaluation model remain fixed throughout the online run and do not participate in this update process.

[0056] Specifically, to quantify the reliability of the main evaluation model under the current input, the system performs a confidence assessment on its output. A comprehensive confidence score C (where C ∈ [0, 1]) is generated to characterize the main evaluation model's grasp of the cell state analysis corresponding to the deeply fused feature vector. The lower the score, the more uncertain the model is about the current prediction result. The confidence score C is calculated based on one or more of the following metrics: a) Measure the confidence score to assess the uncertainty of the prediction. Specifically, if the main evaluation model is a probabilistic model, the prediction variance of its output is used directly. For deterministic deep neural networks, the Monte Carlo method can be used, which involves performing multiple random forward propagations during the inference phase and calculating the statistical variance of the prediction results as an estimate of the uncertainty.

[0057] b) Feature distribution deviation metric confidence score. Specifically, the distribution distance between the current feature and the historical feature dataset used during the training of the main evaluation model is calculated, for example, by calculating the Mahalanobis distance. The larger the distance value, the more the current cell state deviates from the normal operating conditions perceived during model training, and the lower its confidence score accordingly.

[0058] Specifically, the system periodically or after a specific event (such as completing a full charge-discharge cycle) determines whether a preset update trigger condition is met. The trigger condition is designed as a conservative strategy, intended to activate only when there is sufficient evidence that an update is necessary and safe to perform. The trigger condition includes at least one of the following: Condition 1 (Low confidence level corresponds to available truth value): The confidence score C is lower than a first preset threshold (e.g., 0.7), and a high-reliability truth value tag for the current cell status is acquired simultaneously. The truth value tag can be acquired through methods including but not limited to: capacity calibration performed by full charge-discharge cycles combined with ampere-hour integration, standard performance testing in a laboratory environment, or status confirmed by other validated diagnostic methods.

[0059] Condition 2 (Feature Distribution Drift): A significant and stable drift in the statistical distribution of the input deep fusion feature vector is continuously detected by online statistical process control methods (e.g., cumulative sum control charts), indicating that the battery cell may have entered a new aging stage or operating mode.

[0060] The system initiates the parameter update process for the lightweight incremental learning model only when the above triggering conditions are met. This update is limited to the parameters of the lightweight incremental learning model. The parameters of the main evaluation model are strictly isolated and do not participate in the update. To avoid the model overwriting old knowledge when learning new knowledge, this invention employs an anti-forgetting learning algorithm for parameter updates. In a preferred embodiment, an elastic weight solidification algorithm is used, the core process of which includes: For each parameter of the lightweight incremental learning model Calculate its importance weight in past learning tasks. The importance weights This parameter can be approximated by the cumulative squared gradient value during the historical loss function optimization process. The total loss function for this update is then defined. for: in, This refers to the prediction error loss (such as mean squared error) calculated based on new data. For parameters initial value, For parameters The value before this update, λ, is a hyperparameter that balances the preservation of old and new knowledge. The summation term on the right-hand side of the formula constitutes a flexible constraint, which imposes a larger update penalty on parameters with high importance, thereby protecting their encoded old knowledge.

[0061] Minimize the constrained loss function using gradient descent or its variants. Thus, the parameters The process involves updating the network. This process ensures that new knowledge is integrated into the network's redundant parameters or in the direction that minimizes the impact on older knowledge.

[0062] In this application, to enhance the anti-forgetting effect, a fixed-capacity experience replay buffer can be maintained to store a small number of old data samples with historical truth labels. When performing the above update, a portion of the old data can be sampled from this buffer, incorporating its prediction error loss. The calculations enable us to proactively reinforce our memory of key historical patterns.

[0063] S150: Based on the real-time confidence level of the main evaluation model, the first analysis result and the second analysis result are dynamically weighted and fused to obtain the cell analysis result.

[0064] This step adaptively fuses the first and second analysis results generated in step S130 to produce the final cell analysis result. Its core principle is to dynamically adjust the contribution weights of the two results based on the real-time confidence level C of the main evaluation model, thereby achieving an optimal balance between model stability and individual adaptability.

[0065] First, a dynamic fusion weight w is calculated based on the real-time confidence level C of the main evaluation model. The weight w is constructed as a monotonically increasing function of the confidence level C, ensuring that the more confident the main model is in judging the current state, the higher its output will be in the final result.

[0066] In a preferred embodiment, a smooth mapping based on the Sigmoid function is used to calculate the weight w, and the calculation formula is as follows: ); where σ(·) represents the Sigmoid activation function, and k is a positive parameter controlling the slope of the function curve. A preset confidence offset threshold (e.g., 0.5 or 0.6) is used. This is achieved by adjusting the parameters k and... It allows for flexible control over the sensitivity and inflection point of weight changes with confidence level.

[0067] The function is designed such that when the confidence level C is extremely high (e.g., C > 0.9), the weight w approaches 1. In this case, the final analysis result will almost completely adopt the output of the main evaluation model, making full use of its stability and generalization ability trained on massive historical data. When the confidence level C is extremely low (e.g., C < 0.4), the weight w approaches 0. In this case, the final analysis result will mainly rely on the output of the lightweight incremental learning model, because this model has better fitted the individual characteristics of the current battery cell or new operating conditions through online updates.

[0068] In the intermediate confidence interval, the weight w transitions smoothly between 0 and 1, achieving a continuous and non-abrupt adjustment of the contributions of both.

[0069] Based on the calculated dynamic weights The first and second analysis results are weighted and fused to generate the final cell analysis results.

[0070] For regression tasks (such as health status and remaining life estimation), the final cell analysis results are calculated using a linear weighted sum: in, For the final cell analysis results, For dynamic weights, This is the first analysis result. This is the result of the second analysis.

[0071] For classification tasks (such as fault type diagnosis), the first analysis result and the second analysis result are category probability vectors. First, calculate the weighted probability vector: in, Let be the category probability vector of the battery cell. This is the category probability vector corresponding to the first analysis result. This is the category probability vector corresponding to the second analysis result.

[0072] Then, the category with the highest probability is selected as the final classification label, i.e. .

[0073] The final cell analysis results provide the upper-level battery management system with key state information that can be directly used for decision-making. This mainly includes: a health status estimate, which outputs a percentage value (e.g., 82.5%) representing the degree of degradation of the cell's current actual capacity relative to its initial rated capacity—a core indicator for evaluating battery performance; and a remaining service life prediction, which outputs a value and its possible confidence interval (e.g., 450 cycles remaining ± 30 cycles), used to predict the number of cycles or time the cell can still be used before a specific failure threshold, supporting predictive maintenance strategies.

[0074] If the final cell analysis results output a specific classification label (such as "normal", "early lithium plating alarm", "suspected internal short circuit", etc.), then early and qualitative identification of potential faults can be achieved, and graded alarms can be issued based on confidence level and historical trends.

[0075] As can be seen from the above analysis, the battery cell data analysis method based on multi-dimensional fusion provided in this application includes: synchronously preprocessing multi-dimensional heterogeneous raw data from the battery cell; inputting the preprocessed multi-dimensional data into a cross-modal deep feature fusion network to generate a deep fusion feature vector; inputting the deep fusion feature vector into a pre-trained main evaluation model and a lightweight incremental learning model to obtain a first analysis result and a second analysis result, respectively; determining whether to use new data with truth labels to update the parameters of the lightweight incremental learning model based on preset triggering conditions and the confidence evaluation result of the main evaluation model, while keeping the parameters of the main evaluation model unchanged during the update process; and dynamically weighting and fusing the first analysis result and the second analysis result based on the real-time confidence of the main evaluation model to obtain the battery cell analysis result. By using a cross-modal attention mechanism to perform deep feature fusion on the multi-dimensional data of the battery cell, the intrinsic correlation between the multi-dimensional data is fully explored, providing a rich and discriminative information foundation for battery cell status evaluation, and significantly improving the estimation accuracy of key parameters such as battery cell health status and remaining life.

[0076] In one embodiment, the cell data analysis method based on multi-dimensional fusion further includes: The knowledge from the lightweight incremental learning model is periodically distilled into the main evaluation model for incremental updates. Through periodic distillation, the main evaluation model incorporates individualized knowledge validated by the incremental model, progressively optimizing its internal model parameters. This improves the long-term accuracy of the initial analysis results output by the main evaluation model, providing a higher-quality benchmark input for dynamic weighted fusion.

[0077] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0078] Please see Figure 2 , Figure 2 This is a schematic diagram of a cell data analysis device based on multi-dimensional fusion, provided as an embodiment of this application. The cell data analysis device based on multi-dimensional fusion includes modules or units for performing... Figure 1 The steps in the corresponding embodiments. Please refer to the details. Figure 1 The relevant descriptions in the corresponding embodiments are shown below. For ease of explanation, only the parts relevant to this embodiment are shown. See also... Figure 3 A multi-dimensional fusion-based cell data analysis device 200 includes: Data preprocessing module 210 is used to synchronously preprocess multi-dimensional heterogeneous raw data from battery cells; Feature fusion module 220 is used to input preprocessed multi-dimensional data into cross-modal deep feature fusion network to generate deep fusion feature vectors; The collaborative analysis module 230 is used to input the deep fusion feature vector into the pre-trained main evaluation model and the lightweight incremental learning model to obtain the first analysis result and the second analysis result, respectively. The incremental learning module 240 is used to determine whether to use new data with truth labels to update the parameters of the lightweight incremental learning model based on preset triggering conditions and the confidence evaluation result of the main evaluation model. The parameters of the main evaluation model remain unchanged during the update process. The module 250 is used to dynamically weight and fuse the first analysis result and the second analysis result based on the real-time confidence level of the main evaluation model to obtain the cell analysis result.

[0079] In one embodiment, the multi-dimensional heterogeneous raw data includes at least electrical dimension data, thermal dimension data, time dimension data, and operating condition dimension data.

[0080] In one embodiment, the cross-modal deep feature fusion network includes: multiple feature extraction branches adapted to different data dimensions and a cross-modal attention fusion module; the cross-modal attention fusion module is used to dynamically learn and weightedly fuse the features output by each feature extraction branch to generate the deep fusion feature vector.

[0081] In one embodiment, the triggering condition includes at least one of the following: The confidence level of the main evaluation model is lower than the first threshold, and the true value label of the cell state is obtained. Or a significant shift in the feature distribution of the input data is detected.

[0082] In one embodiment, updating the parameters of the lightweight incremental learning model includes: updating the parameters of the lightweight incremental learning model by using elastic weight fixed constraints.

[0083] In one embodiment, the device further includes: The distillation module is used to periodically distill the knowledge of the lightweight incremental learning model into the main evaluation model for progressive updates of the main evaluation model.

[0084] In one embodiment, the cell analysis results are used to indicate the cell's health status, remaining service life, or fault warning information.

[0085] Please see Figure 3 , Figure 3 This is a schematic diagram of a cell data analysis device based on multi-dimensional fusion, provided as an embodiment of this application. Figure 3It is understood that the multi-dimensional fusion-based cell data analysis device 300 includes: a processor 310, a memory 320, and a computer program 330 stored in the memory 320 and executable on the processor 310; when the processor 310 executes the computer program 330, it implements the steps in the above-described embodiments of the multi-dimensional fusion-based cell data analysis method, for example... Figure 1 The steps S110 to S150 are shown. Alternatively, when the processor 310 executes the computer program 330, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 2 The functions of modules 210 to 250 are shown.

[0086] For example, the computer program 330 can be divided into one or more modules / units, one or more of which are stored in the memory 320 and executed by the processor 310 to complete this application. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 330 in a multi-dimensional fusion-based cell data analysis device. For example, the computer program 330 can be divided into a data preprocessing module, a feature fusion module, a collaborative analysis module, an incremental learning module, and an acquisition module. The system includes the following modules: a data preprocessing module for synchronously preprocessing multi-dimensional heterogeneous raw data from battery cells; a feature fusion module for inputting the preprocessed multi-dimensional data into a cross-modal deep feature fusion network to generate a deep fusion feature vector; a collaborative analysis module for inputting the deep fusion feature vector into a pre-trained main evaluation model and a lightweight incremental learning model to obtain a first analysis result and a second analysis result, respectively; an incremental learning module for determining whether to use new data with truth labels to update the parameters of the lightweight incremental learning model based on preset triggering conditions and the confidence evaluation result of the main evaluation model, while the parameters of the main evaluation model remain unchanged during the update process; and an acquisition module for dynamically weighting and fusing the first analysis result and the second analysis result based on the real-time confidence of the main evaluation model to obtain the battery cell analysis result.

[0087] The multi-dimensional fusion-based cell data analysis device 300 provided in this embodiment may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that... Figure 3 This is merely an example of a multi-dimensional fusion-based cell data analysis device 300 and does not constitute a limitation on the multi-dimensional fusion-based cell data analysis device 300. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the multi-dimensional fusion-based cell data analysis device 300 may also include input / output devices, network access devices, buses, etc.

[0088] The processor 310 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0089] The memory 320 can be an internal storage unit of the multi-dimensional fusion-based cell data analysis device 300, such as a hard disk or memory of the multi-dimensional fusion-based cell data analysis device 300. The memory 320 can also be an external storage device of the multi-dimensional fusion-based cell data analysis device 300, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the multi-dimensional fusion-based cell data analysis device 300. Furthermore, the multi-dimensional fusion-based cell data analysis device 300 can include both internal storage units and external storage devices. The memory 320 is used to store computer programs and other programs and data required by the multi-dimensional fusion-based cell data analysis device 300. The memory 320 can also be used to temporarily store data that has been output or will be output.

[0090] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0091] This application also provides a network device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above method embodiments.

[0092] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0093] This application provides a computer program product that, when run on a mobile terminal, enables the mobile terminal to implement the steps described in the above-described method embodiments.

[0094] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0095] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0096] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0097] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or 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; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0098] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0099] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A cell data analysis method based on multi-dimensional data fusion, characterized in that, include: Simultaneous preprocessing of multi-dimensional heterogeneous raw data from battery cells; The preprocessed multi-dimensional data is input into a cross-modal deep feature fusion network to generate a deep fusion feature vector; The deep fusion feature vector is input into the pre-trained main evaluation model and the lightweight incremental learning model to obtain the first analysis result and the second analysis result, respectively. Based on the preset triggering conditions and the confidence evaluation results of the main evaluation model, it is determined whether to use new data with truth labels to update the parameters of the lightweight incremental learning model, and the parameters of the main evaluation model remain unchanged during the update process. Based on the real-time confidence level of the main evaluation model, the first analysis result and the second analysis result are dynamically weighted and fused to obtain the cell analysis result.

2. The method according to claim 1, characterized in that, The multi-dimensional heterogeneous raw data includes at least electrical dimension data, thermal dimension data, time dimension data, and operating condition dimension data.

3. The method according to claim 2, characterized in that, The cross-modal deep feature fusion network includes: multiple feature extraction branches adapted to different data dimensions and a cross-modal attention fusion module; the cross-modal attention fusion module is used to dynamically learn and weightedly fuse the features output by each feature extraction branch to generate the deep fusion feature vector.

4. The method according to claim 1, characterized in that, The triggering condition includes at least one of the following: The confidence level of the main evaluation model is lower than the first threshold, and the true value label of the cell state is obtained. Or a significant shift in the feature distribution of the input data is detected.

5. The method according to claim 4, characterized in that, The parameter update of the lightweight incremental learning model includes: updating the parameters of the lightweight incremental learning model by using elastic weight fixed constraints.

6. The method according to claim 5, characterized in that, The method further includes: The knowledge of the lightweight incremental learning model is periodically distilled into the main evaluation model for incremental updates to the main evaluation model.

7. The method according to claim 1, characterized in that, The cell analysis results are used to indicate the cell's health status, remaining service life, or fault warning information.

8. A cell data analysis device based on multi-dimensional data fusion, characterized in that, include: The data preprocessing module is used to synchronously preprocess multi-dimensional heterogeneous raw data from battery cells; The feature fusion module is used to input preprocessed multi-dimensional data into a cross-modal deep feature fusion network to generate a deep fusion feature vector. The collaborative analysis module is used to input the deep fusion feature vector into the pre-trained main evaluation model and the lightweight incremental learning model to obtain the first analysis result and the second analysis result, respectively. The incremental learning module is used to determine whether to use new data with truth labels to update the parameters of the lightweight incremental learning model based on preset trigger conditions and the confidence evaluation results of the main evaluation model. The parameters of the main evaluation model remain unchanged during the update process. The module is used to dynamically weight and fuse the first analysis result and the second analysis result based on the real-time confidence level of the main evaluation model to obtain the cell analysis result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images