Methods, devices, equipment and storage media for predicting vehicle battery failures

Abstract

This application provides a method, apparatus, device, and storage medium for predicting battery faults in vehicles. The method acquires battery fault prediction data, including global feature data of the battery, multi-source fusion feature data of each battery cell, location data of each battery cell, connection relationship data between each battery cell and other battery cells (excluding itself), and edge feature data of each battery cell. The fault prediction data is input into a fault prediction model, which determines the neighbor perception feature data of each battery cell based on the fault prediction data, and then determines the battery fault prediction result based on the neighbor perception feature data. The fault prediction model of this application can determine the feature data of the battery cells themselves and the correlation between battery cells based on the fault prediction data, thereby effectively capturing the complex nonlinear relationships in the battery system and improving the accuracy of vehicle battery fault prediction.

Description

Technical Field

[0001] This application relates to the field of fault prediction technology, and in particular to a method and apparatus for predicting battery faults in a vehicle, an electronic device, and a storage medium. Background Technology

[0002] Current battery failure prediction technologies are mainly divided into two categories: model-based prediction methods and data-based prediction methods. Model-based prediction methods rely on accurate modeling of the battery's internal physicochemical processes, such as electrochemical impedance spectroscopy models and equivalent circuit models. However, due to the complexity of the battery's internal mechanisms and its susceptibility to multiple factors such as temperature, current, and voltage, modeling is difficult and computationally complex, making accurate prediction of battery failures challenging. Data-based prediction methods learn battery life degradation patterns from historical data to predict battery failures. Common methods include statistical models and machine learning models. However, these methods typically require manual feature extraction, demanding high levels of feature engineering expertise, and have limited capabilities in modeling nonlinear relationships.

[0003] Therefore, how to effectively capture the complex nonlinear relationships in a battery system without relying on precise modeling of the battery's internal physicochemical processes, thereby achieving accurate prediction of battery failures, is a problem that urgently needs to be solved in this field. Summary of the Invention

[0004] This application provides a method for predicting battery failures in vehicles, aiming to solve the problem of how to effectively capture complex nonlinear relationships in the battery system without relying on precise modeling of the battery's internal physicochemical processes, thereby achieving accurate prediction of battery failures.

[0005] Accordingly, embodiments of this application also provide a vehicle battery fault prediction device, an electronic device, and a storage medium to ensure the implementation and application of the above methods.

[0006] To address the aforementioned problems, this application discloses a method for predicting battery failures in a vehicle. The battery comprises multiple battery cells, and the method includes: Obtain fault prediction data for the battery; the fault prediction data includes global feature data of the battery, multi-source fusion feature data of each battery cell, position data of each battery cell, connection relationship data between each battery cell and other battery cells except itself, and edge feature data of each battery cell; the edge feature data is used to characterize the correlation between the battery cell and its neighboring battery cells; The fault prediction data is input into the fault prediction model so that the fault prediction model can determine the neighbor perception feature data of each battery cell based on the fault prediction data, and determine the fault prediction result of the battery based on the neighbor perception feature data.

[0007] Optionally, the methods for obtaining the multi-source fusion feature data of the battery cell include: The vehicle operation data is collected through the first data acquisition branch, the charge and discharge data of each battery cell is collected through the second data acquisition branch, and the electrochemical parameter data of each battery cell is collected through the third data acquisition branch. Data fusion is performed on the vehicle operation data, the charge and discharge data, and the electrochemical parameter data within the same time window to obtain the multi-source fusion feature data of each battery cell.

[0008] Optionally, the data fusion based on the vehicle operation data, each of the charge / discharge data, and each of the electrochemical parameter data within the same time window to obtain multi-source fused feature data for each battery cell includes: The vehicle operation data, the charge / discharge data of each battery cell, and the electrochemical parameter data within the time window are compressed to obtain a summary of the vehicle operation data, a summary of the charge / discharge data of each battery cell, and a summary of the electrochemical parameter data. Based on the summarized vehicle operation data, the summarized charge and discharge data of each battery cell, and the summarized electrochemical parameter data, the multi-source fusion characteristic data of each battery cell are determined.

[0009] Optionally, the charge / discharge data includes the voltage data of the battery cells, and the step of determining the multi-source fusion characteristic data of each battery cell based on the summary vehicle operation data, the summary charge / discharge data of each battery cell, and the summary electrochemical parameter data includes: Based on the voltage data of each battery cell within the time window, the time-domain and frequency-domain characteristics of the voltage of each battery cell within the time window are determined. For each battery cell, the abstract vehicle operation data, the corresponding abstract charge and discharge data, the corresponding abstract electrochemical parameter data, the corresponding voltage time domain features, and the corresponding voltage frequency domain features are spliced ​​together to obtain the multi-source fusion feature data of the battery cell.

[0010] Optionally, the method for obtaining the edge feature data of the battery cell includes: The first difference between the battery cell and the neighboring battery cell is obtained; the first difference includes at least two of the following: temperature difference, current difference, voltage difference, and health status difference. Based on the first difference, a second difference is determined between the battery cell and the neighboring battery cell; The edge feature data of the battery cell is determined based on the first difference and the second difference.

[0011] Optionally, determining the second difference between the battery cell and the neighboring battery cell based on the first difference includes: Obtain the business scenario of the vehicle; The weights corresponding to the difference types of each of the first differences in the business scenario are determined based on the correspondence between the preset business scenario, the preset difference types, and the preset weights. A second difference between the battery cell and its neighboring battery cells is determined based on the first difference and the weights corresponding to the difference types of each first difference.

[0012] Optionally, the fault prediction model is a model trained based on a target loss function, wherein the target loss function is L=a*L1-b*L2, where a and b represent error weight and reward weight, respectively, L1 represents the error between the predicted value and the true value, and L2 represents the reward value calculated based on the current prediction result.

[0013] This application also discloses a vehicle battery fault prediction device, wherein the vehicle battery includes multiple battery cells, and the device includes: The acquisition module is used to acquire fault prediction data of the battery; the fault prediction data includes global feature data of the battery, multi-source fusion feature data of each battery cell, position data of each battery cell, connection relationship data between each battery cell and other battery cells except itself, and edge feature data of each battery cell; the edge feature data is used to characterize the correlation between the battery cell and its neighboring battery cells. The model processing module is used to input the fault prediction data into the fault prediction model, so that the fault prediction model can determine the neighbor perception feature data of each battery cell based on the fault prediction data, and determine the fault prediction result of the battery based on the neighbor perception feature data.

[0014] This application also discloses an electronic device, including: a processor; and a memory storing executable code thereon, which, when executed, causes the processor to perform one or more of the vehicle battery failure prediction methods described in this application.

[0015] This application also discloses a machine-readable medium storing executable code, which, when executed, causes a processor to perform one or more vehicle battery failure prediction methods as described in this application.

[0016] Compared with related technologies, the embodiments of this application have the following advantages: In this embodiment, battery fault prediction data is acquired. This fault prediction data includes global feature data of the battery, multi-source fusion feature data of each battery cell, location data of each battery cell, connection relationship data between each battery cell and other battery cells (excluding itself), and edge feature data of each battery cell. The edge feature data characterizes the correlation between a battery cell and its neighboring battery cells. The fault prediction data is input into a fault prediction model, enabling the model to determine the neighbor perception feature data of each battery cell based on the fault prediction data, and to determine the battery fault prediction result based on the neighbor perception feature data. In this embodiment, the fault prediction data input into the fault prediction model includes global feature data of the battery, multi-source fusion feature data of each battery cell, location data of each battery cell, connection relationship data between each battery cell and other battery cells (excluding itself), and edge feature data of each battery cell. Based on the fault prediction data, the fault prediction model can determine the feature data of the battery cell itself and the correlation between battery cells, thereby effectively capturing the complex nonlinear relationships in the battery system and improving the accuracy of vehicle battery fault prediction. Due to the complexity of the internal mechanism of batteries, they are susceptible to various factors. Fault prediction methods that rely on precise modeling of the internal physicochemical processes of batteries in related technologies are difficult to model and have high computational complexity. The embodiments of this application achieve accurate prediction of battery faults without the need for precise modeling of the complex internal physicochemical processes of batteries, and have lower computational complexity compared to the technical solutions of related technologies. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of an embodiment of a vehicle battery fault prediction method according to this application. Figure 2 This is a fault prediction model architecture diagram of an embodiment of a battery fault prediction method for vehicles according to this application; Figure 3 This is a system architecture diagram of an embodiment of a vehicle battery fault prediction method according to this application; Figure 4 This is a flowchart illustrating a battery fault prediction method for a vehicle according to an embodiment of this application. Figure 5 This is a collaborative deployment diagram of an embodiment of a vehicle battery fault prediction method according to this application; Figure 6 This is a structural block diagram of an embodiment of a vehicle battery fault prediction device according to this application; Figure 7 This is a schematic diagram of the structure of a device provided in an embodiment of this application. Detailed Implementation

[0018] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] Current battery failure prediction technologies are mainly divided into two categories: model-based prediction methods and data-based prediction methods. Model-based prediction methods rely on accurate modeling of the battery's internal physicochemical processes, such as electrochemical impedance spectroscopy models and equivalent circuit models. However, due to the complexity of the battery's internal mechanisms and its susceptibility to multiple factors such as temperature, current, and voltage, modeling is difficult and computationally complex, making accurate prediction of battery failures challenging. Data-based prediction methods learn battery life degradation patterns from historical data to predict battery failures. Common methods include statistical models and machine learning models. However, these methods typically require manual feature extraction, demanding high levels of feature engineering expertise, and have limited capabilities in modeling nonlinear relationships.

[0020] Furthermore, in related AI (Artificial Intelligence) battery fault prediction technologies, most AI prediction systems rely solely on a single type of data (such as battery charge / discharge data) or a limited number of parameters (such as voltage and current) for analysis, failing to effectively integrate multi-dimensional data such as vehicle operating status and battery chemical parameters. For example, traditional monitoring systems primarily assess the health status of vehicle batteries based on conventional parameters such as voltage and current, but these parameters are insufficient to comprehensively reflect the battery's true health condition. Some vehicle batteries may experience problems such as plate sulfation and active material shedding, while the battery's voltage and current remain within the normal range, making it difficult to detect potential faults in a timely manner.

[0021] In summary, the vehicle battery failure prediction methods in related technologies have the following main drawbacks: 1. Limited data dimension: Failure to integrate multi-source battery feature data limits the accuracy of fault prediction; 2. Insufficient model generalization ability: Traditional machine learning methods are difficult to effectively capture the complex nonlinear relationships and dynamic changes of battery systems, especially lacking the ability to model the correlation between battery cells. 3. Weak early fault warning capability: It relies heavily on post-event response or threshold judgment, making it difficult to achieve early fault prediction and warning. 4. Poor environmental adaptability: The impact of different regions and climate conditions on battery performance was not fully considered; 5. Lack of adaptive maintenance decision-making mechanism: Most related solutions adopt fixed thresholds or static strategies, which cannot dynamically adjust maintenance measures according to battery status; 6. Insufficient integration of chemical and physical parameters: Key chemical indicators such as electrolyte concentration and electrode material impedance are not effectively integrated with traditional physical parameters.

[0022] Therefore, how to effectively capture the complex nonlinear relationships in a battery system without relying on precise modeling of the battery's internal physicochemical processes, thereby achieving accurate prediction of battery failures, is a problem that urgently needs to be solved in this field. This application proposes a battery failure prediction method for vehicles to solve, or at least partially solve, the aforementioned technical problems.

[0023] Reference Figure 1 This is a flowchart illustrating the steps of an embodiment of a vehicle battery fault prediction method according to this application, including the following steps: Step 101: Obtain the fault prediction data of the battery; the fault prediction data includes the global feature data of the battery, the multi-source fusion feature data of each battery cell, the position data of each battery cell, the connection relationship data between each battery cell and other battery cells except itself, and the edge feature data of each battery cell; the edge feature data is used to characterize the correlation between the battery cell and its neighboring battery cells.

[0024] The battery in this embodiment includes multiple battery cells. These battery cells are interconnected at their respective physical locations via a predetermined electrical connection method. This electrical connection method may include series connection, parallel connection, busbar connection, or reserved connections.

[0025] In step 101 of this embodiment, fault prediction data corresponding to the battery is obtained. The fault prediction data includes global feature data of the battery, multi-source fusion feature data of each battery cell, position data of each battery cell, connection relationship data between each battery cell and other battery cells excluding itself, and edge feature data of each battery cell.

[0026] The global feature data *g* of the battery represents the overall characteristics of the battery, which contains multiple battery cells; it is also known as the global vector and is used to enable the fault prediction model to understand the global context corresponding to the fault prediction. In one example, it can be used... ^g_dim represents the dimension of the battery's global feature data g, where, Let g be a set of real numbers, and g_dim be the dimension parameter. In one embodiment, the global feature data can be dynamically updated based on the state information of the entire battery.

[0027] The global characteristic data of the battery includes four dimensions: the battery's total voltage, the battery's total current, the vehicle speed corresponding to the battery, and the ambient temperature of the battery. For example, if the battery's total voltage is 38.4V, the battery's total current is 5A, the vehicle speed corresponding to the battery is 60km / h, and the ambient temperature of the battery is 26℃, then the battery's global characteristic data is [38.4V, 5A, 60km / h, 26℃]. Based on this global characteristic data, the fault prediction model can determine that the fault prediction is performed under the global context of a total battery voltage of 38.4V, a total battery current of 5A, a vehicle speed corresponding to the battery of 60km / h, and an ambient temperature of 26℃.

[0028] The multi-source fusion feature data of each battery cell reflects the feature data of each battery cell itself, which is obtained by fusing data from different sources. In one embodiment, the multi-source fusion feature data of each battery cell fuses the vehicle operation data of the corresponding vehicle, the charge and discharge data of the battery cell, the electrochemical parameter data of the battery cell, and the voltage time-domain and voltage frequency-domain features of the battery cell.

[0029] The position data of each battery cell is the physical position of the battery cell in the internal structure of the battery, which can be represented by coordinates (x, y, z).

[0030] In one embodiment, the multi-source fusion feature data of each battery cell can be concatenated with the location data of each battery cell to obtain the node features corresponding to each battery cell. The node feature matrix X of the battery is then obtained based on the node features of multiple battery cells. The fault prediction model can determine the multi-source fusion features and location features corresponding to each battery cell based on the node feature matrix. For example, if the battery contains 12 battery cells, the multi-source fusion feature data of each battery cell contains 24 dimensions, and the location data of each battery cell contains 3 dimensions, then the dimension of the node feature matrix X is... ^(N×d), where N=12 (number of battery cells), and d=27-dimensional node features (24-dimensional multi-source fusion feature data and 3-dimensional location data). For example, one row of the node feature matrix X represents the node features of a battery cell. As an example, it can contain at least the following data [3.2V, 5A, 80% SOC, 25℃, x=10cm, y=20cm, z=5cm, ...]. This example only shows a portion of the node features of a battery cell.

[0031] The connection data between each battery cell and all other battery cells except itself is sparsely represented to indicate the connections between battery cells in the battery. In one embodiment, the connection data between each battery cell and all other battery cells except itself can be represented as an adjacency matrix. For example, for a battery with N battery cells, the adjacency matrix is ​​an N*N matrix. If there is a connection between battery cell i and battery cell j, the element in the i-th row and j-th column of the adjacency matrix is ​​1; otherwise, it is 0. In one embodiment, the adjacency matrix can be pre-fixed based on the physical connection topology of the battery cells.

[0032] The edge feature data of each battery cell characterizes the association between the battery cell and its neighboring battery cells, where neighboring battery cells are those adjacent to the current battery cell. The association between a battery cell and its neighboring battery cells can reflect the differences between the battery cells and their neighboring battery cells. In one embodiment, the edge feature data can be dynamically updated based on the state of the connection relationships between the battery cells themselves.

[0033] Step 102: Input the fault prediction data into the fault prediction model so that the fault prediction model can determine the neighbor perception feature data of each battery cell based on the fault prediction data, and determine the fault prediction result of the battery based on the neighbor perception feature data.

[0034] Reference Figure 2 This is a fault prediction model architecture diagram of an embodiment of a vehicle battery fault prediction method according to this application. In this embodiment, the fault prediction model may include a graph attention network, an encoder (Transformer), a long short-term memory (LSTM) network, an attention mechanism component, and a fault probability prediction head.

[0035] The graph attention network is used to determine the neighbor perception feature data of each battery cell based on fault prediction data. In one embodiment, the target edge feature data corresponding to the edges between the battery cell and each neighboring battery cell is determined. Neighbor feature data is calculated based on the target edge feature data, and then concatenated with the node features of the battery cell itself to obtain the neighbor perception feature data of the battery cell. In one embodiment, the neighbor features include four dimensions: average neighbor temperature difference, total neighbor edge weight, maximum neighbor current difference, and average neighbor SOH difference. Since the node features of the battery cell itself are 27-dimensional, the neighbor perception feature data is 31-dimensional. The neighbor feature data contains information not found in the original node features of the battery cell, but it is crucial in battery fault detection. For example, if a battery cell has a normal voltage but its neighbor average temperature difference is as high as 5°C, the graph attention network can capture the anomaly in the neighbor average temperature difference, allowing subsequent model processing to determine if the battery cell has an overheating risk.

[0036] The encoder takes neighbor-aware feature data as input and captures the dependencies across time, battery, and feature in the neighbor-aware feature data simultaneously through the encoder's multi-head self-attention mechanism, resulting in 128-dimensional global dependency features in time, battery cell, and feature dimensions.

[0037] Specifically, in one example, the neighbor-perceived feature data X' ∈ ^(600×12×31) Input encoder. Where 600 is the time dimension, that is, 600 time steps when the original data is sampled (sampling lasts for 10 minutes, sampling frequency is 1Hz); 12 is the battery cell dimension, representing 12 battery cells; 31 is the feature dimension of the neighbor perception feature data, that is, the neighbor perception feature data corresponding to each battery cell is 31-dimensional data.

[0038] First, the encoder maps the 31-dimensional neighbor perception feature data to a 128-dimensional high-dimensional space, achieving dimensionality upscaling. This is shown in formula (1) below: Z proj = x' · W proj + b proj (1) Where, x' ∈ ³¹ is the 31-dimensional feature vector of a single battery cell in the neighbor perception feature data; W proj ∈ ³¹×¹² 8 , is a learnable weight matrix; b proj ∈ ¹² 8 , is the bias term; Z proj ∈ ¹² 8 It is the result of dimensional elevation.

[0039] Since the encoder itself cannot sense time or spatial location, the upgraded data needs to be encoded so that the model can distinguish different battery cells at different times, such as distinguishing cell No. 3 at the 100th second from cell No. 5 at the 200th second.

[0040] Subsequently, the encoder uses a multi-head self-attention mechanism to determine the correlations between features in the encoded data, outputting a 128-dimensional feature fused from all attention heads. This 128-dimensional feature captures dependencies across time, cells, and features. The number of attention heads can be up to eight, with each head processing 16-dimensional features and focusing on one type of dependency. For example, Head1 focuses on the dependency between voltage and temperature, while Head2 focuses on the dependency between current and state of equilibrium (SOH).

[0041] The encoder transforms the 128-dimensional features output by the multi-head self-attention mechanism into non-linear abstract features through a feedforward neural network. Residual connections and layer normalization operations are performed on the non-linear abstract features to prevent the model from focusing only on abstract features and ignoring the basic data, ultimately resulting in 128-dimensional globally dependent features.

[0042] Therefore, the 128-dimensional global dependency feature in this application embodiment transforms scattered physical parameters into semantic features that can be directly used for fault classification, solving the problem of weak early warning capability caused by insufficient low-dimensional feature expression in related technologies. By further predicting faults based on the 128-dimensional global dependency feature through the fault prediction model, early prediction and warning of faults can be achieved in fault prediction.

[0043] Based on the specific physical meaning of each dimension in the 128-dimensional global dependency features under the long-term trend, LSTM determines the contribution of each dimension to fault prediction under the long-term trend and sorts them. The top 16 dimensions with the highest contribution are selected and retained to obtain 16-dimensional hidden state information.

[0044] The attention mechanism component then determines and weights the attention weights for each feature in the 16 dimensions of hidden state information. For example, if the 16-dimensional hidden state information includes a voltage variance dimension, and the voltage variance is high, indicating a risk of capacity decay or thermal runaway in the battery cell, then a higher attention weight is assigned to the voltage variance. If the 16-dimensional hidden state information includes a temperature gradient dimension, and the temperature gradient is large, indicating a risk of thermal runaway in the battery cell, then a higher attention weight is assigned to the temperature gradient. If the 16-dimensional hidden state information includes a DC internal resistance (DC resistance) dimension, and the DC resistance is increasing, indicating a risk of capacity decay in the battery cell, then a higher attention weight is assigned to the DC resistance. Thus, the first three dimensions of attention weights are voltage variance, temperature gradient, and DC resistance.

[0045] In one embodiment, during the training of the fault prediction model, the attention mechanism component first randomly initializes the attention weights, then calculates the prediction results and loss function through forward propagation, and then backpropagates the loss gradient to the attention layer during the backpropagation stage, thereby updating the attention parameters of the attention mechanism component so that features that contribute more to fault prediction can obtain higher attention weights. Finally, when the model converges, the attention weights tend to stabilize, and at this time the attention weight distribution can effectively reflect the true contribution of each feature to the prediction results.

[0046] In one example, regarding the encoder, LSTM, and attention mechanism components, in a fault prediction scenario of battery capacity degradation, the encoder's role is to discover the correlation between the abnormal electrolyte concentration three months ago and the current increase in internal resistance, thereby establishing a causal relationship across time, across battery cells, and across features, and obtaining 128-dimensional global dependency features; the LSTM's role is to process the 128-dimensional global dependency features from a time window, filtering out instantaneous fluctuations and retaining long-term patterns, obtaining 16-dimensional hidden state information; the attention mechanism component's role is to assign higher attention weights to key dimensions such as internal resistance growth rate and temperature stability from the 16-dimensional hidden state information.

[0047] The weighted hidden state information is input into the fault probability prediction head, and the output fault probability for the preset fault type is used as the fault prediction result.

[0048] The fault probability prediction head in this application embodiment can be a multi-label output layer, employing a multi-label classification architecture, which can simultaneously predict the occurrence probability and remaining service life of multiple fault types in vehicle batteries. When performing fault probability prediction, the fault probability prediction head needs to use global feature data of the battery to provide a global basis for fault judgment, thereby improving the decision-making capability of the fault probability prediction head.

[0049] In one example, the multi-head task of the fault probability prediction head can include six classification prediction tasks and one regression prediction task. The classification prediction tasks output the probability of occurrence for the following preset fault types: capacity decay probability p1, thermal runaway risk probability p2, internal resistance abnormality probability p3, charge / discharge imbalance probability p4, electrolyte concentration abnormality probability p5, and electrode material impedance abnormality probability p6. The regression prediction task outputs the predicted remaining useful life (RUL) of the vehicle battery, normalized and mapped to the [0,1] interval, corresponding to a lifespan range of 0 to 1000 days.

[0050] In one embodiment, the fault prediction model in this application embodiment can also be implemented using a traditional machine learning model (such as XGBoost or random forest), thereby further reducing computational complexity and hardware requirements; or it can be implemented using a hybrid prediction method that combines physical models and data-driven approaches, thereby further improving the interpretability of the model.

[0051] In this embodiment, based on the fault probability for a preset fault type output by the fault prediction model, and combined with a reinforcement learning algorithm, corresponding fault maintenance suggestions can be determined, achieving adaptive maintenance decision-making and solving the problem of the lack of an adaptive maintenance decision-making mechanism in related technologies. The fault maintenance suggestions can be used for dynamic maintenance of the battery pack.

[0052] The preset reinforcement learning algorithm can be the DQN (Deep Q-Network) algorithm. DQN's adaptive maintenance decision-making is a strategy based on deep Q-networks that dynamically adjusts maintenance plans to achieve precise maintenance by monitoring the equipment's operating status in real time. DQN combines the powerful representational capabilities of deep learning with the decision-making capabilities of reinforcement learning, effectively addressing complex situations in equipment maintenance and implementing adaptive maintenance strategies, dynamically adjusting charging parameters, and implementing maintenance measures.

[0053] In one example, the preset fault type, corresponding warning threshold, warning level, and fault maintenance recommendations for the vehicle battery may include the following: (1) Fault type: capacity decay; warning threshold: SOH decrease rate > 5% / month; warning level: yellow; adaptive maintenance measures: dynamically adjust charging cut-off voltage based on reinforcement learning and optimize charging and discharging strategy. (2) Fault type: thermal runaway risk; warning threshold: predicted temperature > 60℃ or temperature rise rate > 5℃ / min; warning level: red; adaptive maintenance measures: activate multi-level cooling strategy and dynamically adjust fan speed and coolant flow.

[0054] (3) Fault type: abnormal internal resistance; warning threshold: internal resistance increase > 20% of the baseline value; warning level: yellow; adaptive maintenance measures: based on the chemical parameter evaluation results, formulate battery balancing strategies or replacement recommendations.

[0055] (4) Fault type: Uneven charging and discharging; Warning threshold: Individual cell voltage difference > 50mV; Warning level: Blue; Adaptive maintenance measures: Activate intelligent balancing algorithm to dynamically adjust the charging current of each individual cell.

[0056] (5) Fault type: Abnormal electrolyte concentration; Warning threshold: Concentration deviates from standard value by ±10%; Warning level: Orange; Adaptive maintenance measures: Adjust charging strategy to avoid deep charging and discharging, and recommend electrolyte testing.

[0057] (6) Fault type: Abnormal electrode material impedance; Warning threshold: Impedance value exceeds threshold by 20%; Warning level: Orange; Adaptive maintenance measures: Optimize charging temperature control and recommend electrode material status detection.

[0058] The fault prediction data input to the fault prediction model in this application embodiment includes global feature data of the battery, multi-source fusion feature data of each battery cell, position data of each battery cell, connection relationship data between each battery cell and other battery cells (excluding itself), and edge feature data of each battery cell. Based on this fault prediction data, the fault prediction model can determine the feature data of the battery cell itself and the correlation between battery cells, thereby effectively capturing the complex nonlinear relationships in the battery system, improving the accuracy of vehicle battery fault prediction, and solving problems such as single data dimension and insufficient model generalization ability in related technologies. Due to the complexity of the internal mechanism of the battery, which is susceptible to various factors, fault prediction methods in related technologies that rely on precise modeling of the battery's internal physicochemical processes are difficult to model and computationally complex. This application embodiment achieves accurate prediction of battery faults without requiring precise modeling of the complex internal physicochemical processes of the battery, and has lower computational complexity compared to related technologies.

[0059] Optionally, the methods for obtaining the multi-source fusion feature data of the battery cell include: The vehicle operation data is collected through the first data acquisition branch, the charge and discharge data of each battery cell is collected through the second data acquisition branch, and the electrochemical parameter data of each battery cell is collected through the third data acquisition branch. Data fusion is performed on the vehicle operation data, the charge and discharge data, and the electrochemical parameter data within the same time window to obtain the multi-source fusion feature data of each battery cell.

[0060] This application embodiment collects data through three different data acquisition branches: the first data acquisition branch collects vehicle operation data, the second data acquisition branch collects charge and discharge data of each battery cell, and the third data acquisition branch collects electrochemical parameter data of each battery cell.

[0061] In one example, the battery charge and discharge data for each battery cell can include six dimensions: cell voltage, bus current, cell temperature, DC internal resistance, SOC (State of Charge) and SOH (State of Health), and the acquisition frequency can be set to 1Hz.

[0062] Vehicle operation data can include six dimensions such as driving speed, driving mode, geographical location (latitude and longitude), ambient temperature and humidity, and charging habits. The collection frequency can be set to 0.1-1Hz, and the collection frequency can be dynamically adjusted by changes in the parameters of the vehicle operation data.

[0063] Battery chemical parameter data can include four dimensions such as electrolyte concentration, electrode material impedance, active material content, and lithium-ion diffusion coefficient, and the acquisition frequency can be set to 0.01Hz.

[0064] In one example, data such as cell voltage, cell temperature, DC internal resistance, SOC, and SOH can be collected for a single battery cell. The operating data of the other battery cells are shared by the battery pack and can be collected for the entire battery or a single battery cell, without the need to collect data for each battery cell individually.

[0065] The collected vehicle operation data, charge and discharge data of each battery cell, and electrochemical parameter data of each battery cell need to be preprocessed. Preprocessing may include data cleaning and data normalization.

[0066] Specifically, data cleaning performs operations such as outlier removal, missing value imputation, and noise smoothing on the raw data of vehicle operation data, charge / discharge data of each battery cell, and electrochemical parameter data of each battery cell. Data normalization uses the min-max normalization method to unify parameters of different dimensions into the [0,1] interval after cleaning. In one example, for a battery containing 12 battery cells, with a time window of T=600s for data acquisition, the tensor shape obtained after normalization of the 16-dimensional battery cell feature data is (T,12,16).

[0067] Data fusion is performed on vehicle operation data, charge / discharge data, and electrochemical parameter data within the same time window to obtain multi-source fused characteristic data for each battery cell. In one example, a time window can be 600 seconds.

[0068] This application integrates vehicle operation data, various charge / discharge data, and electrochemical parameter data for vehicle battery fault prediction. Through multi-source parameter fusion, it improves the accuracy of subsequent vehicle battery fault prediction compared to related technologies, solving problems such as single data dimensions and insufficient integration of chemical and physical parameters in related technologies. Furthermore, the vehicle operation data includes geographical location and environmental temperature and humidity data, addressing the poor environmental adaptability issue in related technologies. By acquiring the temperature and chemical parameter data of individual battery cells and integrating this data into vehicle battery fault prediction, serious safety accidents such as thermal runaway can be effectively prevented, improving vehicle operational safety.

[0069] Optionally, the data fusion based on the vehicle operation data, each of the charge / discharge data, and each of the electrochemical parameter data within the same time window to obtain multi-source fused feature data for each battery cell includes: The vehicle operation data, the charge / discharge data of each battery cell, and the electrochemical parameter data within the time window are compressed to obtain a summary of the vehicle operation data, a summary of the charge / discharge data of each battery cell, and a summary of the electrochemical parameter data. Based on the summarized vehicle operation data, the summarized charge and discharge data of each battery cell, and the summarized electrochemical parameter data, the multi-source fusion characteristic data of each battery cell are determined.

[0070] This application's embodiments compress vehicle operation data, charge / discharge data of each battery cell, and electrochemical parameter data within the same time window to reduce the amount of data processed in subsequent data fusion. In one embodiment, data compression can be applied to the vehicle operation data, charge / discharge data, and electrochemical parameter data corresponding to a single battery cell, where the data corresponding to each feature is averaged over a single time window T (e.g., a 600-second acquisition time window). For example, averaging the data within a single time window T for the individual cell voltage of a single battery cell achieves data compression of the individual cell voltage data.

[0071] After data compression, a 16-dimensional summary vector is obtained for each battery cell, including summary vehicle operation data, summary charge / discharge data for each battery cell, and summary electrochemical parameter data. In one example, with 12 battery cells, the summary vehicle operation data for each cell can be represented as μ_veh = (12×6), the summary charge / discharge data for each cell can be represented as μ_batt = (12×6), and the summary electrochemical parameter data for each cell can be represented as μ_chem = (12×4).

[0072] Based on the abstract vehicle operation data, the abstract charge and discharge data of each battery cell, and the abstract electrochemical parameter data, the multi-source fusion characteristic data of each battery cell were determined.

[0073] The embodiments of this application compress the vehicle operation data, the charge and discharge data of each battery cell, and the electrochemical parameter data within the same time window, which can reduce the amount of data processed by data fusion and reduce the complexity of the data processed by the subsequent fault prediction model.

[0074] Optionally, the charge / discharge data includes the voltage data of the battery cells, and the step of determining the multi-source fusion characteristic data of each battery cell based on the summary vehicle operation data, the summary charge / discharge data of each battery cell, and the summary electrochemical parameter data includes: Based on the voltage data of each battery cell within the time window, the time-domain and frequency-domain characteristics of the voltage of each battery cell within the time window are determined. For each battery cell, the abstract vehicle operation data, the corresponding abstract charge and discharge data, the corresponding abstract electrochemical parameter data, the corresponding voltage time domain features, and the corresponding voltage frequency domain features are spliced ​​together to obtain the multi-source fusion feature data of the battery cell.

[0075] In this embodiment, the voltage time-domain and voltage frequency-domain characteristics of each battery cell within the same time window can also be determined. The voltage data of each battery cell is the single-cell voltage data in the charge-discharge data. For each battery cell, the abstract vehicle operation data, the corresponding abstract charge-discharge data, the corresponding abstract electrochemical parameter data, the corresponding voltage time-domain characteristics, and the corresponding voltage frequency-domain characteristics are spliced ​​together to obtain the multi-source fusion characteristic data of that battery cell.

[0076] In one embodiment, voltage time-domain features and voltage frequency-domain features of the voltage data of each battery cell are extracted using feature engineering methods. The voltage time-domain features may include the mean, variance, and peak value of the individual cell voltage data, while the voltage frequency-domain features may include the spectral energy of the individual cell voltage data. The feature engineering method involves transforming the preprocessed individual cell voltage data into its data features.

[0077] In one example, for a 600s time window of voltage data for a battery cell, there are a total of 600 voltage data points. The time-domain features such as mean, variance, peak value, and RMS value of the 600 voltage data points, as well as the frequency-domain features such as total energy, low-frequency energy, mid-frequency energy, and main frequency ratio, are extracted to obtain 8-dimensional voltage time-domain features and voltage frequency-domain features. The 8-dimensional time-domain features and frequency-domain features are concatenated to the 16-dimensional summary vector of the battery cell (including summary vehicle operation data, corresponding summary charge and discharge data, and corresponding summary electrochemical parameter data) to obtain the 24-dimensional multi-source fusion feature data of the battery cell. For a 12-cell battery, the corresponding 12-cell multi-source fusion feature data matrix X=(600,12,24) is formed.

[0078] Since the voltage data of the battery cell is more sensitive to abnormal changes inside the battery cell, the embodiments of this application determine the corresponding voltage time-domain features and voltage frequency-domain features for each battery cell, and splice them with the abstract vehicle operation data, the corresponding abstract charge and discharge data, and the corresponding abstract electrochemical parameter data. The resulting multi-source fusion feature data can more effectively reflect the state characteristics of the battery cell, thereby improving the accuracy of the model for fault prediction.

[0079] Optionally, the method for obtaining the edge feature data of the battery cell includes: The first difference between the battery cell and the neighboring battery cell is obtained; the first difference includes at least two of the following: temperature difference, current difference, voltage difference, and health status difference. Based on the first difference, a second difference is determined between the battery cell and the neighboring battery cell; The edge feature data of the battery cell is determined based on the first difference and the second difference.

[0080] The embodiments of this application use edge feature data to characterize the correlation between a battery cell and its neighboring battery cells. Specifically, the edge feature data of the correlation between a battery cell and each of its neighboring battery cells may include a first difference and a second difference between the battery cell and its neighboring battery cells.

[0081] The first difference includes at least two of the following: temperature difference, current difference, voltage difference, and health status difference between the battery cell and its neighboring battery cells. A second difference can be determined based on the first difference. In one embodiment, the second difference is the edge weight between the battery cell and its neighboring battery cells. The first and second differences between the battery cell and its neighboring battery cells are used as edge feature data between the battery cell and its neighboring battery cells. This process is repeated to obtain the edge feature data between the battery cell and each of its neighboring battery cells.

[0082] For example, edge feature data can include edge weights, temperature differences between battery cells and neighboring battery cells, current differences between battery cells and neighboring battery cells, and health state differences between battery cells and neighboring battery cells. If a battery contains 12 battery cells, and the edge feature data for each battery cell includes this 4-dimensional data, then the dimension of the edge feature matrix E corresponding to the battery is... ^(M×4), where M is the number of edges connecting the battery cells, and 4 represents the 4-dimensional edge features, including edge weight, temperature difference, current difference, and aging synchronization difference. For example, the edge feature data between a battery cell and one of its neighboring battery cells can be [0.2 (edge ​​weight), 0.5℃ (temperature difference), 0.2A (current difference), 0.1 (SOH difference)].

[0083] In this application embodiment, the first and second differences between the battery cell and its neighboring battery cells are used as edge feature data, which can effectively reflect the differences and correlations between the battery cell and its neighboring battery cells. This allows the graph neural network to capture the complex nonlinear relationships in the battery system based on the fault prediction data, thereby improving the accuracy of vehicle battery fault prediction.

[0084] Optionally, determining the second difference between the battery cell and the neighboring battery cell based on the first difference includes: Obtain the business scenario of the vehicle; The weights corresponding to the difference types of each of the first differences in the business scenario are determined based on the correspondence between the preset business scenario, the preset difference types, and the preset weights. A second difference between the battery cell and the neighboring battery cell is determined based on the first difference and the weights corresponding to the difference types of each first difference.

[0085] In this embodiment, the second difference is calculated based on the first difference between the battery cell and its neighboring battery cells and the weight corresponding to the type of difference of the first difference.

[0086] For each edge between a battery cell and its neighboring battery cells, determine the first difference corresponding to that edge. Normalize this first difference in one dimension to map it to the 0-1 range. Based on a preset business scenario, determine different preset weights for the first differences of different preset difference types. Then, perform a weighted sum of the first differences of different preset difference types to obtain the second difference corresponding to that edge.

[0087] Regarding the correspondence between preset business scenarios, preset difference types, and preset weights, Table 1 below shows some of the possible correspondences in one example. It should be understood that Table 1 in this example only shows some possible correspondences; in actual applications, the correspondences between preset business scenarios, preset difference types, and preset weights can be set according to actual needs.

[0088] Table 1. Correspondence between preset business scenarios, preset difference types, and preset weights.

[0089] In Table 1 of this example, when the vehicle is charging, the risk of thermal runaway and charging anomalies in the vehicle battery require greater attention; therefore, higher weights can be assigned to temperature differences, voltage differences, and current differences. When the vehicle is not charging or driving, the probability of thermal runaway is lower; therefore, higher weights can be assigned to health status differences, while lower weights can be assigned to temperature differences and current differences. It is important to note that the sum of the weights for each difference type should be 1 to avoid numerical overflow.

[0090] In one example, assuming the difference types include temperature difference, current difference, and health status difference, the weighted summation of the second difference can be expressed as the following formula (2): W ij =α normΔT+β normΔI+γ normΔSOH(2) Among them, W ijThe second difference between the i-th and j-th battery cells is α, which is the temperature difference weight (0.5 in the example, the highest priority); normΔT is the normalized temperature difference; β is the current difference weight (0.3 in the example); normΔI is the normalized current difference; γ is the health status difference weight (0.2 in the example); normΔSOH is the normalized health status difference.

[0091] In this application embodiment, a second difference is calculated based on the weight of the first difference between the battery cell and its neighboring battery cells and the difference type of the first difference. In the calculation of the second difference, higher weights are assigned to the difference types that are more critical in the current business scenario, so that the second difference can more effectively reflect the important differences between the battery cell and its neighboring battery cells and improve the accuracy of subsequent fault prediction.

[0092] Optionally, the fault prediction model is a model trained based on a target loss function, wherein the target loss function is L=a*L1-b*L2, where a and b represent error weight and reward weight, respectively, L1 represents the error between the predicted value and the true value, and L2 represents the reward value calculated based on the current prediction result.

[0093] The fault prediction model in this embodiment is trained using a target loss function. The target loss function is L = a*L1 - b*L2, where a and b represent the error weight and reward weight, respectively, L1 represents the error between the predicted value and the true value, and L2 represents the reward value calculated based on the current prediction result.

[0094] In one embodiment, if the current prediction result includes the probability of occurrence corresponding to different preset fault types, the reward value can be determined based on each probability value. Specifically, if the probability of occurrence corresponding to the preset fault type includes capacity decay probability p1, thermal runaway risk probability p2, internal resistance abnormality probability p3, charge-discharge imbalance probability p4, electrolyte concentration abnormality probability p5, and electrode material impedance abnormality probability p6, then the loss function L can be expressed as the following formula (3): L = BCE(p,y) c r_simple(3) Wherein, the error weight is set to 1; BCE(p, y) is the traditional binary cross-entropy loss, used to measure the error between the model's predicted value p and the true value y, ensuring classification accuracy; p = [p1,p2,p3,p4,p5,p6] are the 6 fault probabilities predicted by the model, and y = [y1, y2, y3, y4, y5, y6] are the true fault labels (0 or 1); c∈[0.01,0.10] is an adjustable reward weight, initially set to 0.05 by default, which can be recalibrated during training; the reward function r_simple can be expressed as the following formula (4). c r_simple incorporates the reward function into the loss optimization: r_simple = λ1 p_max + λ2 (1 TV) (4) in, λ1 p_max is the fault risk penalty term; p_max = max(p1,…,p6), which is the highest probability value among the six preset fault types. Physically, it means that the most dangerous part of the battery system determines the overall reliability; λ1 can be set to 0.03 to adjust the penalty intensity; λ2 (1 TV) is the system stability reward; TV is the average first-order difference of the 12-core voltage sequence (normalized to the [0,1] interval), reflecting the voltage fluctuation stability; λ2 can be set to 0.01 to adjust the reward intensity.

[0095] In one embodiment, a reward function can be designed based on the aggregated fault risk features and the rate of change of health status. The aggregated fault risk features can be the prediction results of a fault prediction model, and the rate of change of health status can be determined based on the rate of change of SOH in the battery cell feature data, or based on the average first-order difference of the voltage sequence of the battery pack cells.

[0096] The embodiments of this application train the fault prediction model based on a target loss function that combines error and reward, thereby improving the fault prediction model's ability to capture complex nonlinear relationships in the battery system and enhancing the accuracy of vehicle battery fault prediction.

[0097] To enable those skilled in the art to more clearly understand the vehicle battery failure prediction method shown in the embodiments of this application, the following describes... Figures 3-5 This application describes a method for predicting battery failures in a vehicle, as illustrated in an embodiment.

[0098] Reference Figure 3This is a system architecture diagram of an embodiment of a vehicle battery fault prediction method according to this application.

[0099] This application embodiment employs a cloud-edge-device collaborative prediction and maintenance system architecture to implement a vehicle battery fault prediction method. Data is collected and initially processed at the terminal layer before being sent to the edge layer. The edge layer, located on the terminal, deploys a fault prediction model that filters the data submitted by the terminal layer in real time and uses the model to predict faults. The cloud layer is used for big data storage and deep analysis, optimizing the model parameters of the fault prediction model to achieve an organic combination of real-time early warning and long-term optimization. The collaboration layer enables cross-system data sharing and decision support.

[0100] The edge layer collects fault prediction data and uploads it to the cloud. The cloud then uses the data submitted by the edge layer to retrain or fine-tune the fault prediction model to optimize its parameters. The cloud then distributes the optimized model parameters back to the edge layer, thereby updating and optimizing the parameters of the fault prediction model deployed on the edge layer.

[0101] Reference Figure 4 This is a flowchart illustrating a battery fault prediction method for a vehicle according to an embodiment of this application.

[0102] like Figure 4 As shown, firstly, multi-source data is collected from the vehicle battery to obtain raw data for fault prediction. This raw data can be obtained from multiple channels, including databases, APIs (Application Programming Interfaces), and sensors. Data cleaning operations are then performed on the raw data, including outlier removal, missing value imputation, and noise smoothing. The cleaned data is then normalized using either min-max standardization to unify parameters of different dimensions to the [0,1] interval, or Z-score standardization. Feature engineering is then performed on the preprocessed data, including extracting, filtering, and combining key features to integrate structured and unstructured data, achieving data fusion. Finally, the processed data is input into a machine learning model or deep learning model (i.e., a fault prediction model) to obtain the vehicle battery fault prediction results and corresponding fault maintenance suggestions.

[0103] Reference Figure 5 This is a collaborative deployment diagram of an embodiment of a vehicle battery fault prediction method according to this application.

[0104] The cloud-edge-device collaborative prediction and maintenance system architecture of this application's embodiments involves the terminal layer collecting vehicle battery data from devices and uploading it to the edge layer. A fault prediction model is deployed on the edge layer to process the data submitted by the terminal layer and achieve fault prediction. The cloud layer globally optimizes the model parameters of the fault prediction model and distributes the updated fault prediction model to achieve model parameter synchronization. In one embodiment, a federated learning architecture can be used instead of the cloud-edge-device architecture to enhance data privacy protection.

[0105] The fault prediction data input to the fault prediction model in this application embodiment includes global feature data of the battery, multi-source fusion feature data of each battery cell, position data of each battery cell, connection relationship data between each battery cell and other battery cells (excluding itself), and edge feature data of each battery cell. Based on this fault prediction data, the fault prediction model can determine the feature data of the battery cell itself and the correlation between battery cells, thereby effectively capturing the complex nonlinear relationships in the battery system and improving the accuracy of vehicle battery fault prediction. Due to the complexity of the battery's internal mechanisms and its susceptibility to various factors, fault prediction methods in related technologies that rely on precise modeling of the battery's internal physicochemical processes are difficult to model and computationally complex. This application embodiment achieves accurate prediction of battery faults without requiring precise modeling of the complex internal physicochemical processes of the battery, resulting in lower computational complexity compared to related technologies.

[0106] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of this application.

[0107] Based on the above embodiments, this embodiment also provides a vehicle battery fault prediction device, which can be applied to electronic devices such as terminal devices and servers.

[0108] Reference Figure 6 The diagram shows a structural block diagram of an embodiment of a vehicle battery fault prediction device according to this application, which may specifically include the following modules: The acquisition module 601 is used to acquire fault prediction data of the battery; the fault prediction data includes global feature data of the battery, multi-source fusion feature data of each battery cell, position data of each battery cell, connection relationship data between each battery cell and other battery cells except itself, and edge feature data of each battery cell; the edge feature data is used to characterize the correlation between the battery cell and its neighboring battery cells. The model processing module 602 is used to input the fault prediction data into the fault prediction model, so that the fault prediction model can determine the neighbor perception feature data of each battery cell based on the fault prediction data, and determine the fault prediction result of the battery based on the neighbor perception feature data.

[0109] Optionally, the device includes: The data acquisition module is used to acquire vehicle operation data of the vehicle through the first data acquisition branch, acquire charge and discharge data of each battery cell through the second data acquisition branch, and acquire electrochemical parameter data of each battery cell through the third data acquisition branch. The data fusion module is used to fuse the vehicle operation data, the charge and discharge data, and the electrochemical parameter data within the same time window to obtain multi-source fused feature data of each battery cell.

[0110] Optionally, the data fusion module includes: The data compression submodule is used to compress the vehicle operation data, the charge and discharge data of each battery cell and the electrochemical parameter data within the time window to obtain the vehicle operation data summary, the battery cell charge and discharge data summary and the electrochemical parameter data summary. The data fusion determination submodule is used to determine the multi-source fusion characteristic data of each battery cell based on the abstract vehicle operation data, the abstract charge and discharge data of each battery cell, and the abstract electrochemical parameter data.

[0111] Optionally, the charge / discharge data includes the voltage data of the battery cell, and the fused data determination submodule is specifically used for: Based on the voltage data of each battery cell within the time window, the time-domain and frequency-domain characteristics of the voltage of each battery cell within the time window are determined. For each battery cell, the abstract vehicle operation data, the corresponding abstract charge and discharge data, the corresponding abstract electrochemical parameter data, the corresponding voltage time domain features, and the corresponding voltage frequency domain features are spliced ​​together to obtain the multi-source fusion feature data of the battery cell.

[0112] Optionally, the device includes: The first difference acquisition module is used to acquire a first difference between the battery cell and the neighboring battery cell; the first difference includes at least two of the following: temperature difference, current difference, voltage difference, and health status difference. The second difference acquisition module is used to determine a second difference between the battery cell and the neighboring battery cell based on the first difference. An edge feature determination module is used to determine the edge feature data of the battery cell based on the first difference and the second difference.

[0113] Optionally, the second difference acquisition module includes: The scenario determination submodule is used to obtain the business scenario of the vehicle; The weight determination submodule is used to determine the weight corresponding to each of the first differences in the business scenario based on the correspondence between the preset business scenario, the preset difference type and the preset weight; The second difference determination submodule is used to determine the second difference between the battery cell and the neighboring battery cell based on the first difference and the weights corresponding to the difference types of each first difference.

[0114] Optionally, the fault prediction model is a model trained based on a target loss function, wherein the target loss function is L=a*L1-b*L2, where a and b represent error weight and reward weight, respectively, L1 represents the error between the predicted value and the true value, and L2 represents the reward value calculated based on the current prediction result.

[0115] This application also provides a non-volatile readable storage medium storing one or more modules (programs). When these modules are applied to a device, they enable the device to execute the instructions for the method steps in this application.

[0116] This application provides one or more machine-readable media storing instructions that, when executed by one or more processors, cause an electronic device to perform one or more of the methods described in the above embodiments. In this application, the electronic device includes various types of devices such as terminal devices and servers (clusters).

[0117] The embodiments of this disclosure can be implemented as an apparatus configured as desired using any suitable hardware, firmware, software, or any combination thereof, including electronic devices such as terminal devices, servers (clusters), etc. Figure 7 An exemplary apparatus 700 is schematically shown that can be used to implement the various embodiments described in this application.

[0118] In one embodiment, Figure 7 An exemplary device 700 is shown, which includes one or more processors 702, a control module (chipset) 704 coupled to at least one of the processors 702, a memory 706 coupled to the control module 704, a non-volatile memory (NVM) / storage device 708 coupled to the control module 704, one or more input / output devices 710 coupled to the control module 704, and a network interface 712 coupled to the control module 704.

[0119] Processor 702 may include one or more single-core or multi-core processors, and processor 702 may include any combination of general-purpose processors or special-purpose processors (e.g., graphics processors, application processors, baseband processors, etc.). In some embodiments, device 700 can serve as a terminal device, server (cluster), or other device as described in the embodiments of this application.

[0120] In some embodiments, apparatus 700 may include one or more computer-readable media (e.g., memory 706 or NVM / storage device 708) having instructions 714 and one or more processors 702 that are combined with the one or more computer-readable media and configured to execute the instructions 714 to implement the module and thus perform the actions described in this disclosure.

[0121] In one embodiment, the control module 704 may include any suitable interface controller to provide any suitable interface to at least one of the processors 702 and / or any suitable device or component communicating with the control module 704.

[0122] The control module 704 may include a memory controller module to provide an interface to the memory 706. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0123] Memory 706 may be used, for example, to load and store data and / or instructions 714 for device 700. In one embodiment, memory 706 may include any suitable volatile memory, such as suitable DRAM. In some embodiments, memory 706 may include double data rate type quad synchronous dynamic random access memory (DDR4 SDRAM).

[0124] In one embodiment, the control module 704 may include one or more input / output controllers to provide an interface to the NVM / storage device 708 and (one or more) input / output devices 710.

[0125] For example, NVM / storage device 708 may be used to store data and / or instructions 714. NVM / storage device 708 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable (one or more) non-volatile storage devices (e.g., one or more hard disk drive (HDD), one or more optical disc (CD) drives, and / or one or more digital universal optical disc (DVD) drives).

[0126] NVM / storage device 708 may include storage resources that are physically part of a device on which device 700 is mounted, or that are accessible to the device but do not necessarily have to be part of the device. For example, NVM / storage device 708 may be accessed via a network via one or more input / output devices 710.

[0127] One or more input / output devices 710 may provide an interface for device 700 to communicate with any other suitable device. Input / output devices 710 may include communication components, audio components, sensor components, etc. A network interface 712 may provide an interface for device 700 to communicate via one or more networks. Device 700 may wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, such as accessing a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof.

[0128] In one embodiment, at least one of the processors 702 may be logically packaged with one or more controllers (e.g., memory controller modules) of the control module 704. In one embodiment, at least one of the processors 702 may be logically packaged with one or more controllers of the control module 704 to form a system-in-package (SiP). In one embodiment, at least one of the processors 702 may be integrated with the logic of one or more controllers of the control module 704 on the same die. In one embodiment, at least one of the processors 702 may be integrated with the logic of one or more controllers of the control module 704 on the same die to form a system-on-a-chip (SoC).

[0129] In various embodiments, device 700 may be, but is not limited to, a server, desktop computing device, or mobile computing device (e.g., laptop computing device, handheld computing device, tablet computer, netbook, etc.). In various embodiments, device 700 may have more or fewer components and / or different architectures. For example, in some embodiments, device 700 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touchscreen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0130] The detection device can use a main control chip as a processor or control module, and sensor data, position information, etc. can be stored in a memory or NVM / storage device. The sensor group can be used as an input / output device, and the communication interface can include a network interface.

[0131] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0132] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0133] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable vehicle battery fault prediction terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable vehicle battery fault prediction terminal device, generate instructions for implementing the process... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0134] These computer program instructions may also be stored in a computer-readable storage medium that can direct a fault prediction terminal device of a computer or other programmable vehicle battery to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0135] These computer program instructions can also be loaded onto a computer or other programmable vehicle battery fault prediction terminal device, causing a series of operational steps to be executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0136] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0137] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0138] The above provides a detailed description of a vehicle battery fault prediction method and apparatus, an electronic device, and a storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and its core ideas. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting battery failures in a vehicle, characterized in that, The battery includes multiple battery cells, and the method includes: Obtain fault prediction data for the battery; the fault prediction data includes global feature data of the battery, multi-source fusion feature data of each battery cell, position data of each battery cell, connection relationship data between each battery cell and other battery cells except itself, and edge feature data of each battery cell; the edge feature data is used to characterize the correlation between the battery cell and its neighboring battery cells; The fault prediction data is input into the fault prediction model so that the fault prediction model can determine the neighbor perception feature data of each battery cell based on the fault prediction data, and determine the fault prediction result of the battery based on the neighbor perception feature data.

2. The vehicle battery failure prediction method according to claim 1, characterized in that, The methods for obtaining the multi-source fusion feature data of the battery cell include: The vehicle operation data is collected through the first data acquisition branch, the charge and discharge data of each battery cell is collected through the second data acquisition branch, and the electrochemical parameter data of each battery cell is collected through the third data acquisition branch. Data fusion is performed on the vehicle operation data, the charge and discharge data, and the electrochemical parameter data within the same time window to obtain the multi-source fusion feature data of each battery cell.

3. The vehicle battery failure prediction method according to claim 2, characterized in that, The data fusion based on the vehicle operation data, the charge / discharge data, and the electrochemical parameter data within the same time window yields multi-source fused feature data for each battery cell, including: The vehicle operation data, the charge / discharge data of each battery cell, and the electrochemical parameter data within the time window are compressed to obtain a summary of the vehicle operation data, a summary of the charge / discharge data of each battery cell, and a summary of the electrochemical parameter data. Based on the summarized vehicle operation data, the summarized charge and discharge data of each battery cell, and the summarized electrochemical parameter data, the multi-source fusion characteristic data of each battery cell are determined.

4. The vehicle battery fault prediction method according to claim 3, characterized in that, The charge / discharge data includes the voltage data of the battery cells. The determination of multi-source fusion characteristic data for each battery cell based on the summarized vehicle operation data, the summarized charge / discharge data of each battery cell, and the summarized electrochemical parameter data includes: Based on the voltage data of each battery cell within the time window, the time-domain and frequency-domain characteristics of the voltage of each battery cell within the time window are determined. For each battery cell, the abstract vehicle operation data, the corresponding abstract charge and discharge data, the corresponding abstract electrochemical parameter data, the corresponding voltage time domain features, and the corresponding voltage frequency domain features are spliced ​​together to obtain the multi-source fusion feature data of the battery cell.

5. The vehicle battery fault prediction method according to claim 1, characterized in that, The methods for obtaining the edge feature data of the battery cell include: The first difference between the battery cell and the neighboring battery cell is obtained; the first difference includes at least two of the following: temperature difference, current difference, voltage difference, and health status difference. Based on the first difference, a second difference is determined between the battery cell and the neighboring battery cell; The edge feature data of the battery cell is determined based on the first difference and the second difference.

6. The vehicle battery failure prediction method according to claim 5, characterized in that, Determining the second difference between the battery cell and the neighboring battery cell based on the first difference includes: Obtain the business scenario of the vehicle; The weights corresponding to the difference types of each of the first differences in the business scenario are determined based on the correspondence between the preset business scenario, the preset difference types, and the preset weights. A second difference between the battery cell and its neighboring battery cells is determined based on the first difference and the weights corresponding to the difference types of each first difference.

7. The vehicle battery failure prediction method according to claim 1, characterized in that, The fault prediction model is a model trained based on a target loss function, which is L=a*L1-b*L2, where a and b represent error weight and reward weight, respectively, L1 represents the error between the predicted value and the true value, and L2 represents the reward value calculated based on the current prediction result.

8. A battery fault prediction device for a vehicle, characterized in that, The vehicle battery includes multiple battery cells, and the device includes: The acquisition module is used to acquire fault prediction data of the battery; the fault prediction data includes global feature data of the battery, multi-source fusion feature data of each battery cell, position data of each battery cell, connection relationship data between each battery cell and other battery cells except itself, and edge feature data of each battery cell; the edge feature data is used to characterize the correlation between the battery cell and its neighboring battery cells. The model processing module is used to input the fault prediction data into the fault prediction model, so that the fault prediction model can determine the neighbor perception feature data of each battery cell based on the fault prediction data, and determine the fault prediction result of the battery based on the neighbor perception feature data.

9. An electronic device, characterized in that, include: processor; and A memory having executable code stored thereon, which, when executed, causes the processor to perform the battery failure prediction method for a vehicle as described in any one of claims 1-7.

10. A machine-readable medium having executable code stored thereon, which, when executed, causes a processor to perform a battery failure prediction method for a vehicle as claimed in any one of claims 1-7.

Images