Vision feature driven battery system consistency differentiation anomaly diagnosis method and system

Abstract

The application provides a visual feature driven battery system consistency differentiation anomaly diagnosis method, system and computer program product. The method forms a state space plane with all single battery voltage averages at any sampling time as the horizontal axis and the deviation of each single battery voltage from the average as the vertical axis; forms a state distribution matrix through gridding and frequency statistics, and constructs a two-dimensional probability density distribution image; uses the two-dimensional probability density distribution image data to train a deep visual feature diagnosis network to realize accurate and reliable diagnosis of battery system consistency differentiation anomalies. The application effectively retains the global consistency distribution information of the battery system through explicit visual representation of the battery system consistency state, and realizes unified feature expression between battery systems of different series scales using a normalization mechanism, thereby improving the engineering adaptability and cross-topology generalization ability of the battery system consistency differentiation anomaly diagnosis.

Description

Technical Field

[0001] This invention relates to the field of intelligent state monitoring and anomaly diagnosis technology for power batteries, and more specifically, to a visual feature-driven method and system for diagnosing battery system consistency differentiation anomalies. Background Technology

[0002] Driven by the rapid development of new energy vehicles and new power systems, lithium-ion batteries, as the core energy storage unit, are evolving towards higher voltage and larger scale. As the series connection scale of battery systems continues to expand, with the number of individual cells reaching hundreds or more, parameter differences between cells caused by multi-physics coupling during manufacturing, assembly, and service gradually accumulate and amplify. This leads to a sharp reduction in the usable depth of discharge and energy of the battery system, exacerbates internal electrothermal imbalances, and may even induce safety accidents. Therefore, conducting accurate consistency anomaly diagnosis for large-scale battery systems has become an urgent need to ensure the safe, reliable, and cost-effective operation of large-scale battery energy storage systems.

[0003] Currently, diagnostic methods for battery system inconsistencies mainly include model-driven and data-driven approaches. Model-driven methods are typically based on electrochemical models or equivalent circuit models, achieving anomaly diagnosis by estimating individual cell parameters online and combining the residuals between acquired measurements and the model. While these methods have clear physical meaning, their modeling accuracy is highly dependent on the accuracy of model parameter identification, and the computational complexity increases significantly as the scale of battery systems grows, making it difficult to meet the needs of rapid online diagnosis for large-scale battery systems.

[0004] Data-driven methods primarily rely on time-series operational data collected by the battery management system. They extract statistical features such as voltage standard deviation, information entropy, and characteristic parameters for specific voltage ranges, and then combine these with machine learning models to diagnose anomalies. However, existing data-driven methods typically compress high-dimensional time-series data into a small number of low-order statistical features, resulting in a loss of global distribution information and making it difficult to effectively reflect the complex consistency differentiation anomalies within the battery system. Especially for large-scale battery systems, as the system size increases exponentially, under long-term imbalance conditions, the individual cell parameters often evolve from a single Gaussian distribution to a bimodal or even multimodal Gaussian distribution. Such differentiation anomalies are not obvious at the terminal voltage, making it difficult for traditional diagnostic methods based on low-dimensional statistical features to effectively identify them. Furthermore, existing data-driven methods are topology-dependent; the non-equilibrium evolution of individual cell parameters is strongly bound to the series-parallel topology of the battery pack. When the series scale of the battery system changes, the dimension of the input features changes, making it difficult to directly transfer and use the original model, and lacking cross-topology generalization ability.

[0005] Therefore, there is an urgent need for a battery system consistency differentiation anomaly diagnosis method that can preserve global consistency distribution information and has cross-topology generalization ability, so as to improve the accuracy of battery system consistency differentiation anomaly identification and the versatility of engineering deployment. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, this invention proposes a visual feature-driven method and system for diagnosing battery system consistency differentiation anomalies. By converting the original high-dimensional variable-length time-series data into a two-dimensional probability density distribution image, data structure alignment is achieved through visual representation. Combined with a deep residual network model, high-precision anomaly diagnosis is completed. This transforms the traditional complex time-series anomaly diagnosis problem into a highly robust image pattern recognition problem. While preserving global consistency distribution information, it does not depend on the series-parallel topology of the battery pack, making it particularly suitable for diagnosing consistency differentiation anomalies in large-scale battery systems.

[0007] To achieve the above objectives, in a first aspect, the present invention provides a visual feature-driven method for diagnosing battery system consistency differentiation anomalies, comprising the following steps:

[0008] S1: Construct a state space plane with the average value of all individual cell voltages of the battery system at any sampling time as the horizontal axis and the deviation of each individual cell voltage from the average value as the vertical axis.

[0009] S2: Perform two-dimensional discretization on the state space plane, traverse the time-series voltage sampling points of the battery system under specific operating conditions, count the cumulative frequency of the voltage trajectory in each grid cell, and form a state distribution matrix.

[0010] S3: Perform column-independent normalization on the state distribution matrix to construct a two-dimensional probability density distribution image to achieve feature space scale alignment between battery systems of different topological scales;

[0011] S4: Construct a deep visual feature diagnostic network model, taking the two-dimensional probability density distribution image after the original acquired voltage data is converted as input, and outputting a binary classification diagnostic result of normal consistency or abnormal differentiation; train the deep visual feature diagnostic network model using the normal consistency dataset and the abnormal differentiation state consistency dataset;

[0012] S5: Based on the trained deep visual feature diagnostic network model, diagnose whether there is a consistency differentiation anomaly in the battery system and output the anomaly diagnosis result.

[0013] The above technical solution maps high-dimensional variable-length time-series running data into a two-dimensional probability density distribution image, and combines a column-independent normalization mechanism to achieve cross-topological feature space alignment. It also uses a deep convolutional neural network model to extract consistent visual features with physical semantics, thereby achieving accurate identification of battery system consistency differentiation anomalies.

[0014] In some embodiments, in step S2, the state space plane is further divided into a two-dimensional grid with fixed precision, wherein the maximum deviation of the individual unit voltage is within the range of [-V]. bound ,+V bound ]wait V diff The interval is discretized into H voltage ranges; the average cell voltage is referenced to the charge / discharge cutoff voltage covering [V]. min V max ] Scope, etc. V avg The voltage intervals are discretized into W voltage intervals, resulting in a two-dimensional discretized grid of H×W.

[0015] In some embodiments, the column independent normalization formula in step S3 is as follows:

[0016]

[0017] Where M(x,y) is the cumulative frequency of the grid cell (x,y). Here, H represents the normalized probability density, and H is the number of grid cells in the voltage deviation direction.

[0018] In some embodiments, the deep visual feature diagnostic network model employs a ResNet series network structure. It is understood that the deep visual feature diagnostic network model can also be other deep learning network models capable of classifying and recognizing two-dimensional images.

[0019] In some embodiments, the deep visual feature diagnosis network model takes a two-dimensional probability density distribution image as input, passes through a convolutional layer, a max pooling layer, a residual feature extraction layer, and a global average pooling layer in sequence, and finally outputs a binary classification result by a fully connected layer.

[0020] In some embodiments, a Rint equivalent circuit model is used to describe the response of individual cells under constant voltage and current conditions within the system. A Monte Carlo method is used to randomly sample within the consistency parameter space, orthogonally generating system-level simulated voltage time-series datasets with different levels of consistency. Simulation datasets containing both normal consistency and divergence anomalies are constructed as the training and validation sets for the deep visual feature diagnostic network model. In this way, divergence anomalies in real battery systems can be simulated without a large number of measured data samples, effectively expanding the training dataset of the network model.

[0021] In some embodiments, consistent normal samples are generated using a multivariate Gaussian distribution with parameters following:

[0022]

[0023] Where θ is the individual cell consistency parameter vector, which follows a Gaussian distribution N, Q is the cell capacity, SOC0 is the initial state of charge of the cell, R is the internal resistance of the cell, and μ is the mean of the multivariate Gaussian distribution of the individual cell consistency parameters. Its standard deviation;

[0024] The anomalous samples with consistent differentiation were generated using a binormal Gaussian mixture distribution with parameters following:

[0025]

[0026] in, Let N be the distribution description of the consistency parameter vector φ, where N is a Gaussian distribution, and μ1 and μ2 are the two means of the binormal mixture Gaussian distribution after the differentiation of the individual consistency parameters. and Let α be its standard deviation, and let α control the relative size of the binormal Gaussian mixture distribution, with a value range of [0,1].

[0027] In this way, the differentiation anomalies of real battery systems can be simulated without a large number of measured data samples, which can effectively expand the training dataset of the network model.

[0028] In a second aspect, the present invention provides a visual feature-driven battery system consistency differentiation anomaly diagnosis system, used to perform the battery system consistency differentiation anomaly diagnosis method as described above, including:

[0029] The data acquisition module is used to acquire the timing data of the individual cell voltages in the battery system.

[0030] The state space plane construction module is used to construct the state space plane for two-dimensional discretized mesh processing;

[0031] The visual representation module is used to construct a two-dimensional probability density distribution image;

[0032] The model training module is used to train a deep visual feature diagnostic network model.

[0033] The anomaly diagnosis module is used to perform consistency differentiation anomaly diagnosis and identification;

[0034] The results output module is used to output the diagnostic results of the consistency differentiation anomaly.

[0035] In some embodiments, the data acquisition module includes a simulation voltage time series dataset generation module.

[0036] Finally, this application provides a computer program product that, when run on a processor, executes the steps of the battery system consistency differentiation anomaly diagnosis method as described above.

[0037] Compared to existing traditional methods based on low-order statistics or time series modeling, the beneficial effects of this invention include: effectively preserving the consistent distribution characteristics of large-scale battery systems, enabling automatic identification and accurate diagnosis of battery system consistency differentiation anomalies; and being directly applicable to new topological battery systems with the same material system without any retraining or structural adjustment, possessing engineering generalization potential for deployment in large-scale, multi-configuration battery systems. Attached Figure Description

[0038] The invention, its features and advantages will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0039] Figure 1 This is a flowchart of a visual feature-driven method for diagnosing battery system consistency differentiation anomalies in an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the single-cell voltage response curves of the battery system in the normal consistency state (a) and abnormal differentiation state (b) in an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of a two-dimensional discretized mesh in an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the two-dimensional probability density distribution images corresponding to the normal consistency state (a) and abnormal differentiation state (b) of the battery system in an embodiment of the present invention;

[0043] Figure 5 This is a schematic diagram of the deep visual feature diagnosis network model structure in an embodiment of the present invention;

[0044] Figure 6 This is a schematic diagram of a visual feature-driven battery system consistency differentiation anomaly diagnosis system in an embodiment of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but these are not intended to limit the scope of the invention.

[0046] In the following detailed description, numerous specific details are set forth to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that well-known algorithms and models are not shown in detail to avoid obscuring the gist of the invention; and that the techniques not detailed in the following effect examples are readily available prior art.

[0047] Example

[0048] See Figure 1 This embodiment provides a visual feature-driven method for diagnosing battery system consistency differentiation anomalies, specifically including the following steps:

[0049] S0: The effects of this invention's embodiments are introduced by constructing a battery system simulation voltage dataset with controllable consistency parameters. In practical application scenarios, data acquisition can directly obtain the voltage acquisition data of all individual cells in the battery system through the data acquisition module. For example, a Rint equivalent circuit model is used to describe the response of the individual cell voltages under constant current conditions within the system. Combined with the Monte Carlo method, random sampling is performed within the consistency parameter space to orthogonally generate system-level simulation voltage time series datasets with different consistency levels.

[0050] Furthermore, step S0 includes: the inconsistency between cells in the battery system is simulated by assigning differentiated values ​​to the cell's capacity (Q), initial state of charge (SOC0), and internal resistance (R) to construct normal consistency samples and abnormal differentiation state consistency samples.

[0051] Normally consistent samples are generated using a multivariate Gaussian distribution with parameters following:

[0052]

[0053] Where θ is the individual cell consistency parameter vector, which follows a Gaussian distribution N, Q is the cell capacity, SOC0 is the initial state of charge of the cell, R is the internal resistance of the cell, and μ is the mean of the multivariate Gaussian distribution of the individual cell consistency parameters. Its standard deviation;

[0054] Abnormally differentiated samples were generated using a double normal mixture Gaussian distribution, with parameters following:

[0055]

[0056] in, Let N be the distribution description of the consistency parameter vector φ, where N is a Gaussian distribution, and μ1 and μ2 are the two means of the binormal mixture Gaussian distribution after the differentiation of the individual consistency parameters. and Let α be its standard deviation, and let α control the relative size of the binormal Gaussian mixture distribution, with a value range of [0,1].

[0057] Based on the Rint equivalent circuit model, the voltage response of all individual cells in the simulated battery system under constant current conditions is as follows:

[0058]

[0059]

[0060]

[0061] Among them, R i (t) represents the ohmic resistance of monomer i at time t, R base (SOC i (t) is the internal resistance-SOC reference curve. R i V is the internal resistance bias of the single cell i. i (t) represents the lower terminal voltage of single cell i at time t, U OCV Open circuit voltage, SOC i (t) represents the state of charge of individual cell i at time t, I represents the constant operating current of the system (positive when discharging), and SOC. 0,i Let Q be the charge state of monomer i at the initial time t0. i Let i be the capacity of a single cell.

[0062] S1: Construct a state space plane with the average value of all individual cell voltages of the battery system at any sampling time as the horizontal axis and the deviation of each individual cell voltage from the average value as the vertical axis.

[0063]

[0064]

[0065] In the formula, V is the average voltage of all individual cells in the system at time t. i (t) represents the voltage of the i-th cell at time t, ΔV i (t) represents the relative deviation of the individual cell voltage, and n represents the number of cells connected in series.

[0066] The schematic diagrams of the single-cell voltage response curves of the battery system provided in this embodiment under normal consistency and abnormal differentiation states are shown below. Figure 2 As shown, when the battery system's consistency state differs, the voltage characteristics between individual cells in the normal consistency state and the abnormal differentiation state are not significantly different. Therefore, when diagnosing directly using low-order statistics based on raw voltage time-series data, it is difficult to effectively identify large-scale battery system consistency differentiation anomalies. This indicates that explicit encoding and visual representation of voltage distribution patterns are important means to achieve high-precision anomaly diagnosis.

[0067] S2: The inconsistent state space plane is discretized by gridding, and the time-series voltage sampling points of the battery system under specific operating conditions are traversed. The cumulative frequency of the voltage trajectory in each grid cell is counted to form a state distribution matrix.

[0068] Further, step S2 includes: further dividing the state space plane into a two-dimensional grid with fixed precision. Specifically, for this embodiment, the maximum range of individual cell voltage deviation is [-0.08]. , +0.08]V, discretized into 40 voltage intervals at 0.004V intervals; average single-cell voltage reference charge / discharge cutoff voltage coverage range [2.76] , 4.20] V is discretized into 48 voltage intervals at 0.03V intervals, resulting in a two-dimensional discretized grid as shown below. Figure 3 As shown.

[0069] S3: Perform column-independent normalization on the state distribution matrix to construct a two-dimensional probability density distribution image to achieve feature space scale alignment between battery systems of different topological scales;

[0070] Furthermore, step S3 includes: independently normalizing the statistical frequency column:

[0071]

[0072] Where M(x,y) is the cumulative frequency of the grid cell (x,y). The normalized probability density is represented by H, which is the number of grid cells in the voltage deviation direction. In this embodiment, the value is 40.

[0073] After normalization, the state distribution matrix is ​​visualized to obtain a fixed-size two-dimensional probability density distribution image. A schematic diagram of the two-dimensional probability density distribution images corresponding to the normal consistency state and abnormal differentiation state of the battery system provided in this embodiment of the invention is shown below. Figure 4 As shown, when the battery system exhibits good consistency, the voltage probability distributions of all cells are highly similar and concentrated within a certain voltage range, appearing as a relatively narrow and horizontal bright main band along the zero point of the vertical axis. However, when the system consistency shows abnormal differentiation, the statistical cumulative frequency splits towards different deviation voltage ranges, resulting in multiple distinct bright bands in the image. Therefore, the two-dimensional probability density distribution image characterizes the battery system consistency state with distinguishable visual features.

[0074] S4: Construct a deep visual feature diagnostic network model, and train the deep visual feature diagnostic network model based on the two-dimensional probability density distribution image samples after the original collected voltage data is converted.

[0075] A deep visual feature diagnostic network model is constructed using the ResNet series network architecture, including convolutional layers, max pooling layers, residual feature extraction layers, global average pooling layers, and classification output layers.

[0076] S5: Based on the trained deep visual feature diagnostic network model, diagnose whether there is a consistency differentiation anomaly in the battery system and output the anomaly diagnosis result.

[0077] This embodiment provides a schematic diagram of a deep visual feature diagnostic network model structure, as shown below. Figure 5 As shown, this network model takes a two-dimensional probability density distribution image as input, and passes through convolutional layers, max pooling layers, four sets of residual feature extraction layers, and a global average pooling layer in sequence. Finally, a fully connected layer outputs the binary classification result.

[0078] In this embodiment, a battery system simulation dataset containing both normal and abnormal data was constructed for validation. 2.5% of the full simulation dataset was used as the training set, and the remaining 97.5% was used as the validation set. Abnormal data was classified as positive, and the overall accuracy, precision, recall, and F1 score were calculated using a confusion matrix to quantitatively evaluate the diagnostic model's performance. The diagnostic comparison results with traditional time-series methods based on raw data, such as the two-layer Long Short-Term Memory (LSTM) network model and the one-dimensional convolutional neural network (1D-CNN) model, are shown in Table 1.

[0079] Table 1 Performance index results of different anomaly diagnosis models

[0080]

[0081] Traditional temporal modeling methods, such as two-layer LSTM and 1D-CNN, use raw time-series data collected by the battery management system as model input. In contrast, the visual-driven method of this invention converts this time-series data into a two-dimensional probability density distribution image, which is then used as the model input. All comparative models use the presence or absence of consistency differentiation anomalies as their model output.

[0082] The comparative two-layer LSTM time series model consists of two cascaded long short-term memory network models, with 128 hidden units in each layer. Since the data sequence lengths of the discharge process vary under different consistency states of the battery system, the input time series data are uniformly truncated to the same fixed length before being input into the model. The LSTM model dynamically models long-term dependencies in the time series through input gates, forget gates, and output gates. The network model introduces a dropout mechanism between layers to reduce the risk of overfitting, and completes the classification output of normal and abnormal consistency through fully connected layers.

[0083] The comparative 1D-CNN model consists of two convolutional layers, a global average pooling layer, and a fully connected classification layer. The input temporal data is also truncated before being fed into the model. The first convolutional layer of the 1D-CNN model extracts basic local dynamic features, while the second convolutional layer further extracts higher-order temporal pattern features. Each convolutional layer is followed by a batch normalization layer, a ReLU activation function, and a max-pooling layer. Finally, a global average pooling layer compresses the temporal features, and a fully connected layer outputs a consistency anomaly diagnostic result.

[0084] As shown in Table 1, the F1 score of the visual feature-driven consistency differentiation anomaly diagnosis method reached 92.2%, which improved by 4.4 percentage points and 26.2 percentage points respectively compared with the temporal modeling 1D-CNN method and the two-layer LSTM method, achieving the optimal balance between recall and precision.

[0085] The cross-topological zero-shot generalization performance of the visual feature-driven consistency differentiation anomaly diagnosis method was verified by constructing multiple sets of simulation data of battery systems with different series connections. The specific steps are as follows:

[0086] A deep visual feature diagnostic network model, trained and converged using less than 3% of the full sample data from a 120-cell battery system, was directly applied to simulation datasets of battery systems with different series connection counts for consistency differentiation anomaly diagnosis without any parameter fine-tuning. The construction method for battery system datasets in cross-topology scenarios is exactly the same as for the 120-cell dataset, except that the total number of individual cells connected in series is set to 200 and 300, while all other consistency parameters are set consistently with the 120-cell battery system dataset.

[0087] Table 2 Performance metrics of the deep visual feature diagnostic network model on battery systems with different tandem numbers.

[0088]

[0089] As shown in Table 2, when faced with data sources with a surge in dimensionality, the anomaly detection accuracy of this model remains stable at over 93%, and the recall and precision even show a certain degree of improvement. There is no performance degradation of the model that is common in cross-topology applications.

[0090] As described above, this invention provides a visual feature-driven method for diagnosing battery system consistency differentiation anomalies. By constructing a state space plane and performing grid-based discretization, the time-series voltage sampling points of the battery system under specific operating conditions are traversed, and the cumulative frequency of the voltage trajectory in each grid cell is statistically analyzed to form a state distribution matrix. Column-independent normalization is used to lock the relative probability distribution, achieving scale invariance and alignment of the feature space. This significantly reduces the learning difficulty of the model in high-dimensional feature spaces, improves the sensitivity and diagnostic robustness of identifying potential differentiation anomalies in large-scale battery systems under extremely small sample conditions, provides model support for diagnosing consistency differentiation anomalies in large-scale battery systems, and has significant engineering application value. It alleviates the technical problems of low model diagnostic accuracy and poor cross-topology generalization in existing technologies.

[0091] This invention provides a visual feature-driven method and system for diagnosing battery system consistency differentiation anomalies. Figure 6 As shown. The system includes:

[0092] Data acquisition module 10 is used to acquire timing data of individual cell voltages in the battery system;

[0093] State space plane construction module 20 is used to construct the state space plane for two-dimensional discretized mesh processing;

[0094] Visual representation module 30 is used to construct a two-dimensional probability density distribution image;

[0095] Model training module 40 is used to train a deep visual feature diagnostic network model;

[0096] The anomaly diagnosis module 50 performs consistency differentiation anomaly diagnosis and identification based on the trained deep visual feature diagnosis network model;

[0097] The result output module 60 is used to output the diagnostic results of the consistency differentiation anomaly.

[0098] The modules work together to achieve automated and accurate diagnosis of consistency differentiation anomalies across topologies in large-scale battery systems.

[0099] The present invention also provides a computer program product, which, when run on a processor, executes the steps of the battery system consistency differentiation anomaly diagnosis method as described above.

[0100] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0101] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0102] 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 implementations should not be considered beyond the scope of this invention.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, 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.

[0105] 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.

[0106] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0107] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above. Systems and structures not described in detail should be understood as being implemented in a conventional manner in the art. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the present invention. This does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention are still within the scope of protection of the present invention.

Claims

1. A visual feature-driven method for diagnosing battery system consistency differentiation anomalies, characterized in that: Includes the following steps: S1: Construct a state space plane with the average value of all individual cell voltages of the battery system at any sampling time as the horizontal axis and the deviation of each individual cell voltage from the average value as the vertical axis; S2: Perform two-dimensional discretization on the state space plane, traverse the time-series voltage sampling points of the battery system under specific operating conditions, count the cumulative frequency of the voltage trajectory in each grid cell, and form a state distribution matrix. S3: Perform column-independent normalization on the state distribution matrix to construct a two-dimensional probability density distribution image to achieve feature space scale alignment between battery systems of different topological scales; S4: Construct a deep visual feature diagnostic network model, taking the two-dimensional probability density distribution image after the original acquired voltage data is converted as input, and outputting a binary classification diagnostic result of normal consistency or abnormal differentiation; train the deep visual feature diagnostic network model using the normal consistency dataset and the abnormal differentiation state consistency dataset; S5: Based on the trained deep visual feature diagnostic network model, diagnose whether there is a consistency differentiation anomaly in the battery system and output the anomaly diagnosis result.

2. The visual feature-driven method for diagnosing battery system consistency differentiation anomalies according to claim 1, characterized in that, In step S2, the state space plane is further divided into a two-dimensional grid with fixed precision, wherein the maximum deviation of the individual unit voltage is within the range of [-V]. bound +V bound ]wait V diff The interval is discretized into H voltage ranges; the average cell voltage is referenced to the charge / discharge cutoff voltage covering [V]. min V max ] Scope, etc. V avg The voltage intervals are discretized into W voltage intervals, resulting in a two-dimensional discretized grid of H×W.

3. A visual feature-driven method for diagnosing battery system consistency differentiation anomalies according to claim 1 or 2, characterized in that, The column independent normalization formula in step S3 is as follows: Where M(x,y) is the cumulative frequency of the grid cell (x,y). Here, H represents the normalized probability density, and H is the number of grid cells in the voltage deviation direction.

4. A visual feature-driven method for diagnosing battery system consistency differentiation anomalies according to claim 1 or 2, characterized in that, The deep visual feature diagnostic network model adopts the ResNet series network structure.

5. The visual feature-driven method for diagnosing battery system consistency differentiation anomalies according to claim 4, characterized in that, The deep visual feature diagnosis network model takes a two-dimensional probability density distribution image as input, and passes through a convolutional layer, a max pooling layer, a residual feature extraction layer, and a global average pooling layer in sequence. Finally, a fully connected layer outputs the binary classification result.

6. A visual feature-driven method for diagnosing battery system consistency differentiation anomalies according to claim 1 or 2, characterized in that, The Rint equivalent circuit model is used to describe the response of the individual units under constant voltage and current conditions within the system. The Monte Carlo method is used to randomly sample in the consistency parameter space to orthogonally generate system-level simulation voltage time series datasets with different consistency levels. Simulation datasets containing normal consistency and abnormal differentiation are constructed as the training and validation sets of the deep visual feature diagnosis network model.

7. The visual feature-driven method for diagnosing battery system consistency differentiation anomalies according to claim 6, characterized in that, Consistent normal samples are generated using a multivariate Gaussian distribution with parameters following: Where θ is the individual cell consistency parameter vector, which follows a Gaussian distribution N, Q is the cell capacity, SOC0 is the initial state of charge of the cell, R is the internal resistance of the cell, and μ is the mean of the multivariate Gaussian distribution of the individual cell consistency parameters. Its standard deviation; The anomalous samples with consistent differentiation were generated using a binormal Gaussian mixture distribution with parameters following: in, Let N be the distribution description of the consistency parameter vector φ, where N is a Gaussian distribution, and μ1 and μ2 are the two means of the binormal mixture Gaussian distribution after the differentiation of the individual consistency parameters. and Let α be its standard deviation, and let α control the relative size of the binormal Gaussian mixture distribution, with a value range of [0,1].

8. A visual feature-driven diagnostic system for battery system consistency differentiation anomalies, characterized in that, A method for performing the battery system consistency differentiation anomaly diagnosis method as described in any one of claims 1 to 7, comprising: The data acquisition module is used to acquire the timing data of the individual cell voltages in the battery system. The state space plane construction module is used to construct the state space plane for two-dimensional discretized mesh processing; The visual representation module is used to construct a two-dimensional probability density distribution image; The model training module is used to train a deep visual feature diagnostic network model. The anomaly diagnosis module is used to perform consistency differentiation anomaly diagnosis and identification; The results output module is used to output the diagnostic results of the consistency differentiation anomaly.

9. A visual feature-driven battery system consistency differentiation anomaly diagnostic system according to claim 8, characterized in that, The data acquisition module includes a simulation voltage time series dataset generation module.

10. A computer program product, characterized in that, When the program runs on the processor, it performs the steps of the battery system consistency differentiation anomaly diagnosis method as described in any one of claims 1 to 7.

Images