Battery pack temperature state detection method, system, device, and storage medium

Abstract

This application discloses a method, system, device, and storage medium for detecting the temperature state of a battery pack, belonging to the field of battery technology. The method includes: determining a first current residual for the battery cell and a second current residual for the battery pack based on the current temperature data of the battery cell, the current temperature statistics of the battery pack, and a target detection model; wherein the target detection model is trained based on an autoencoder and normal state data of the battery pack; determining a first target anomaly threshold for the battery cell and a second target anomaly threshold for the battery pack based on a preset operating condition correction strategy, a first preset anomaly threshold, and a second preset anomaly threshold; determining a temperature anomaly in the battery cell if the first current residual is greater than the first target anomaly threshold; and determining a temperature anomaly in the battery pack if the second current residual is greater than the second target anomaly threshold. This application enables accurate detection of the temperature state of the battery pack and each battery cell.

Description

Technical Field

[0001] This application relates to the field of battery technology, specifically to battery pack temperature state detection methods, systems, devices, and storage media. Background Technology

[0002] With the widespread use of power batteries in electric vehicles and energy storage systems, battery management systems typically rely on thermistors to sample and set fixed absolute temperature or temperature rise slope thresholds to determine the battery's temperature safety status. However, the thresholds of this method are derived from empirical calibration under laboratory conditions and fail to fully account for the drastic fluctuations in ambient temperature, high-current charging and discharging, and heat dissipation conditions under real-world road conditions, thus leading to inaccurate battery temperature status detection. Summary of the Invention

[0003] A method, system, device, and storage medium for detecting the temperature status of a battery pack are provided to achieve accurate detection of the temperature status of the battery pack.

[0004] Firstly, a method for detecting the temperature state of a battery pack is provided, the battery pack comprising multiple battery cells; the method includes: Based on the current temperature data of the battery cell, the current temperature statistics of the battery pack, and the target detection model, the first current residual of the battery cell and the second current residual of the battery pack are determined; wherein, the target detection model is trained based on the autoencoder and the normal state data of the battery pack. Based on the preset operating condition correction strategy, the first preset abnormal threshold, and the second preset abnormal threshold, the first target abnormal threshold of the battery cell and the second target abnormal threshold of the battery pack are determined. If the current residual is greater than the first target anomaly threshold, the cell temperature is determined to be abnormal. If the second current residual is greater than the second target anomaly threshold, the battery pack temperature is determined to be abnormal.

[0005] In some embodiments, normal state data includes historical temperature data of the battery cell, historical temperature statistics of the battery pack, and operating condition information of the battery pack; the self-encoder includes a first self-encoder branch, a second self-encoder branch, a latent space, and a decoder; the method for determining the target detection model includes: Establish the first self-encoder branch for the battery cell and the second self-encoder branch for the battery pack; Based on historical temperature data, historical temperature statistics, operating condition information, the first self-encoder branch, and the second self-encoder branch, the cell representation vector of the battery cell and the whole pack representation vector of the battery pack are determined. Based on the potential space, cell representation vector, and package representation vector, a first fused representation vector and a second fused representation vector are determined; wherein, the first fused representation vector is the cell representation vector after fusing the package representation vector; and the second fused representation vector is the package representation vector after fusing the individual cell representation vectors. Based on the first fusion representation vector, the second fusion representation vector, the decoder, historical temperature data, and historical temperature statistics data, the first reconstruction residual and the second reconstruction residual are determined. The target detection model is obtained by training the first and second reconstruction residuals using a preset loss function.

[0006] In some embodiments, the first autoencoder branch includes a temporal convolutional network; the second autoencoder branch includes a gated recurrent unit; and the cell representation vector of the battery cell and the overall battery pack representation vector are determined based on historical temperature data, historical temperature statistics data, operating condition information, the first autoencoder branch, and the second autoencoder branch, including: The cell representation vector is determined based on historical temperature data, operating condition information, and a temporal convolutional network. Based on historical temperature statistics, operating condition information, and gated loop units, the whole package representation vector is determined.

[0007] In some embodiments, determining the first reconstruction residual and the second reconstruction residual based on the first fusion representation vector, the second fusion representation vector, the decoder, historical temperature data, and historical temperature statistics data includes: The first fused representation vector and the second fused representation vector are input into the decoder to obtain the cell temperature reconstruction data and the whole pack statistics reconstruction data; The first reconstruction residual is determined based on cell temperature reconstruction data and historical temperature data; The second reconstruction residual is determined based on the reconstructed data of the whole package statistics and the historical temperature statistics.

[0008] In some embodiments, the method further includes: determining a first preset anomaly threshold based on the mean and standard deviation of all first reconstruction residuals of the battery cell within a first preset time period; The second preset anomaly threshold is determined based on the mean and standard deviation of all second reconstruction residuals of the battery pack within the first preset time period.

[0009] In some embodiments, the method further includes: determining the mean of a first residual and the standard deviation of a first residual based on the first current residual of the battery cell at each time step within a second preset time period; Based on the second current residual of the battery pack at each time step within the second preset time period, determine the mean of the second residual and the standard deviation of the second residual; If the mean and standard deviation of the first residual, and / or the mean and standard deviation of the second residual, meet the preset alarm conditions, an early warning or graded alarm will be issued.

[0010] In some embodiments, determining a first target anomaly threshold for the battery cell and a second target anomaly threshold for the battery pack based on a preset operating condition correction strategy, a first preset anomaly threshold, and a second preset anomaly threshold includes: The operating condition correction factor is determined based on the current operating condition of the battery pack; the current operating condition is either an extreme condition or a non-extreme condition. The first target anomaly threshold is determined based on the operating condition correction factor and the first preset anomaly threshold. The second target anomaly threshold is determined based on the operating condition correction factor and the second preset anomaly threshold.

[0011] Secondly, embodiments of this application also provide a battery pack temperature status detection system, the battery pack including multiple battery cells, the system including: The first determining module is used to determine the first current residual of the battery cell and the second current residual of the battery pack based on the current temperature data of the battery cell, the current temperature statistics data of the battery pack, and the target detection model; wherein, the target detection model is trained based on the autoencoder and the normal state data of the battery pack. The second determining module is used to determine the first target abnormal threshold of the battery cell and the second target abnormal threshold of the battery pack based on the preset working condition correction strategy, the first preset abnormal threshold and the second preset abnormal threshold. The third determining module is used to determine the cell temperature abnormality when the first current residual is greater than the first target abnormality threshold. The fourth determination module is used to determine the battery pack temperature anomaly when the second current residual is greater than the second target anomaly threshold.

[0012] Thirdly, embodiments of this application also provide an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the computer program, when executed by the processor, implements the method described in the first aspect.

[0013] Fourthly, embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps of the method described in the first aspect.

[0014] Beneficial Effects: This application provides a battery pack temperature state detection method, system, device, and storage medium. The battery pack temperature state detection method includes: determining a first current residual of the battery cell and a second current residual of the battery pack based on the current temperature data of the battery cell, the current temperature statistics of the battery pack, and a target detection model; wherein, the target detection model is trained based on an autoencoder and normal state data of the battery pack; determining a first target anomaly threshold of the battery cell and a second target anomaly threshold of the battery pack based on a preset operating condition correction strategy, a first preset anomaly threshold, and a second preset anomaly threshold; determining that the battery cell temperature is abnormal if the first current residual is greater than the first target anomaly threshold; and determining that the battery pack temperature is abnormal if the second current residual is greater than the second target anomaly threshold. The battery pack temperature state detection method provided in this application outputs the first current residual of the battery cell and the second current residual of the battery pack through a target detection model trained based on an autoencoder and normal state data of the battery pack. It determines whether the temperature of the battery cell is abnormal by comparing the first current residual with the first target anomaly threshold, and determines whether the temperature of the battery pack is abnormal by comparing the second current residual with the second target anomaly threshold, thereby achieving accurate detection of the temperature state of the battery pack and each battery cell. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a battery pack temperature state detection method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the self-encoder provided in the embodiments of this application; Figure 3 This is a flowchart illustrating the early warning and graded warning processes provided in the embodiments of this application; Figure 4 This is a schematic diagram of the overall process of a battery pack temperature status detection method provided in the embodiments of this application; Figure 5 This is a schematic diagram of the principle structure of a battery pack temperature status detection system provided in the embodiments of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] "A and / or B" includes the following three combinations: A only, B only, and a combination of A and B.

[0020] The use of "applies to" or "configured to" in this application implies open and inclusive language, which does not exclude the applicability to or configuration to devices performing additional tasks or steps. Additionally, the use of "based on" implies openness and inclusivity, because processes, steps, calculations, or other actions "based on" one or more of the stated conditions or values ​​may in practice be based on additional conditions or values ​​beyond those stated.

[0021] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0022] The applicant's research revealed that, with the widespread use of power batteries in electric vehicles and energy storage systems, battery management systems typically rely on thermistors to sample and set fixed absolute temperature or temperature rise slope thresholds to determine safety status. These thresholds, derived from empirical calibration under laboratory conditions, fail to adequately account for the drastic fluctuations in ambient temperature, high-current charging and discharging, and heat dissipation under real-world road conditions. Furthermore, real-world temperature anomalies are relatively rare, making it difficult to provide sufficient and diverse fault labels for supervised learning models. Therefore, in the absence of fault labels, accurate monitoring of the battery pack's temperature status is particularly crucial.

[0023] Related technologies employ a fixed threshold scheme, simplifying temperature safety to a single condition: "exceeding a threshold indicates an anomaly." When ambient temperature rises or current surges, the normal thermal rise will fall within the threshold range and trigger an alarm. To reduce false alarms, the threshold is often forced to be raised. However, once the threshold is raised, rapid temperature rise caused by internal short circuits or localized heat dissipation obstruction may be masked below the threshold, ultimately leading to missed detections. When using supervised learning models to alleviate this contradiction, the scarcity of fault samples makes it difficult to train a classifier with reliable generalization ability. Therefore, these technologies fail to achieve a balance between sensitivity, false alarm rate, and adaptability.

[0024] In view of this, embodiments of this application provide a method, system, device, and storage medium for detecting the temperature state of a battery pack. Embodiments of this application use a target detection model trained based on an autoencoder and normal state data of the battery pack to output a first current residual of the battery cell and a second current residual of the battery pack. The temperature of the battery cell is determined to be abnormal by comparing the first current residual with a first target abnormality threshold, and the temperature of the battery pack is determined to be abnormal by comparing the second current residual with a second target abnormality threshold. This achieves accurate detection of the temperature state of the battery pack and each battery cell.

[0025] Figure 1 This is a flowchart illustrating a battery pack temperature state detection method provided in an embodiment of this application. This application provides a battery pack temperature state detection method applicable to a battery management system, enabling accurate detection of the temperature state of the battery pack and individual cells. This method can be executed by a battery pack temperature state detection system, which can be implemented in software and / or hardware and can be configured in the processor or controller of the battery management system. Please refer to... Figure 1 The method includes the following steps: Step 110: Based on the current temperature data of the battery cell, the current temperature statistics of the battery pack, and the target detection model, determine the first current residual of the battery cell and the second current residual of the battery pack.

[0026] The current temperature data of each cell in the battery pack can be obtained in real time through temperature sensors. The current temperature statistics of the battery pack include the current average temperature, the current highest cell temperature, the current lowest cell temperature, the current maximum temperature difference (i.e., the difference between the highest and lowest cell temperatures), the current temperature standard deviation, the current average temperature rise rate, and the current maximum temperature rise rate.

[0027] The target detection model is trained based on autoencoder and normal state data of the battery pack.

[0028] The method for obtaining normal state data includes: collecting cell temperature, package temperature statistics, and operating condition information under various operating scenarios, and constructing a dataset containing only normal states by slicing the data into fixed time windows. For data with abnormal states, iterative self-cleaning can be performed during the target detection model training phase; that is, samples with high training loss are removed during training, and training continues until the training loss meets the requirements.

[0029] The operating condition information includes current, voltage, state of charge (SOC), state of health (SOH), ambient temperature, thermal management signals, and coolant temperature. Overall package temperature statistics include average temperature. Maximum cell temperature Minimum cell temperature Maximum temperature difference (i.e., the difference between the highest and lowest cell temperatures) Temperature standard deviation Average temperature rise rate and the highest temperature rise rate .

[0030] Specifically, by inputting the current temperature data of each battery cell and the current temperature statistics of the battery pack into the trained target detection model, the first current residual of each battery cell and the second current residual of the battery pack can be directly output. Since the target detection model is trained based on the autoencoder and the normal state data of the battery pack, its detection accuracy is relatively high. Therefore, it can improve the accuracy of the first and second current residuals, which is beneficial for accurately judging the temperature state of each battery cell and the battery pack based on the first and second current residuals, thereby realizing multi-scale temperature anomaly detection at the battery pack and individual battery cell levels.

[0031] Figure 2 This is a schematic diagram of the self-encoder provided in an embodiment of this application. In some embodiments, normal state data includes historical temperature data of the battery cell, historical temperature statistics of the battery pack, and operating condition information of the battery pack. See also... Figure 2The autoencoder comprises a first autoencoder branch (including a temporal convolutional network), a second autoencoder branch (including gated recurrent units), a latent space (attention mechanism), and a decoder; wherein, the transposed temporal convolutional network yields the decoder corresponding to the first autoencoder branch, and the gated recurrent units constitute the decoder corresponding to the second autoencoder branch. The method for determining the object detection model includes the following steps: Step 1: Establish the first self-encoder branch of the battery cell and the second self-encoder branch of the battery pack.

[0032] Among them, the first autoencoder branch is a single cell autoencoder branch that shares parameters among the individual cells. It takes the temperature sequence of the individual cell (i.e., the temperature data of the individual cell) and operating condition information as input to learn the normal evolution law of each individual cell.

[0033] The second self-encoder branch is a whole-pack encoder branch, which takes the whole-pack temperature statistics and operating condition information as input to learn the normal mode of the overall thermal characteristics of the battery pack.

[0034] Step 2: Based on historical temperature data, historical temperature statistics, operating condition information, the first autoencoder branch, and the second autoencoder branch, determine the cell representation vector of the battery cell and the overall battery pack representation vector.

[0035] The historical temperature statistics of the battery pack include the historical average temperature, the historical highest cell temperature, the historical lowest cell temperature, the historical maximum temperature difference (i.e., the difference between the highest and lowest cell temperatures), the historical temperature standard deviation, the historical average temperature rise rate, and the historical maximum temperature rise rate.

[0036] In some embodiments, the first autoencoder branch includes a temporal convolutional network; the second autoencoder branch includes a gated recurrent unit; determining the cell representation vector of the battery cell and the overall battery pack representation vector based on historical temperature data, historical temperature statistics data, operating condition information, the first autoencoder branch, and the second autoencoder branch includes: determining the cell representation vector based on historical temperature data, operating condition information, and the temporal convolutional network; and determining the overall battery pack representation vector based on historical temperature statistics data, operating condition information, and the gated recurrent unit.

[0037] The autoencoder structure of a single battery cell adopts a temporal convolutional network (TCN), and the network weights are shared among the cells.

[0038] The input to the autoencoder of a single battery cell is the historical temperature sequence of the single cell (i.e., historical temperature data of the cell) and the operating condition information sequence. The autoencoder models are independent but share parameters among the sequences. The temporal convolutional network (TCN) mainly consists of multi-channel convolutional layers and residual blocks. The multi-channel convolutional layers are used to extract local features in the sequence, and the single-layer convolution includes causal convolution and dilated convolution. Among them, causal convolution is used to ensure that the output only accepts input from historical moments, while dilated convolution expands the receptive field so that the network can learn more information. The single-layer convolution is shown in the following formula: ; in, Indicates the number of input channels; Indicates the kernel length; Indicates the first The output channel is for the first... Each input channel in displacement Weight at each location; Indicates the first The bias of each output channel; This indicates the porosity and determines the skip sampling interval; Indicates the first Each input channel at time The value; Indicates the first Each output channel at time The output of .

[0039] The receptive field of multi-layer dilated convolution is calculated as follows: ; in, This indicates the number of historical time steps that the top-level output unit can obtain; Indicates the number of floors; Indicates the first The void ratio of a layer is commonly used. .

[0040] Residual layers are used to facilitate deep network training and improve model expressive power, while avoiding the gradient vanishing problem. The calculation of the residual layer is shown in the following formula: ; ; ; in, Indicates the first Layer input; This represents the entire convolution operator; Presentation layer normalization operation; This represents a non-linear activation function, typically ReLU; This represents the normalized convolution result; This represents the intermediate features after activation.

[0041] Top-level output The cell vector representation of a single battery cell is as follows: ; in, Indicates the first Convolution and residual connection calculations for layers.

[0042] The structure of the whole packet autoencoder adopts a gated recurrent unit (GRU), and uses the hidden state output at the last time step as the representation vector of the whole packet.

[0043] The calculation process for the recurrent neural unit is shown in the following formula: ; ; ; ; in, express The input at any given time is a vector composed of the above-mentioned total package statistics and operating condition information; express The hidden state at any given moment; , , This indicates the update of the gate's weight and bias; This represents a non-linear activation function, typically ReLU; express The gate output is updated continuously; , , This indicates the weight and bias of the reset gate; express The output of the reset gate at any time; , , This represents the weights and biases used when calculating candidate hidden states; express Candidate hidden states at time t; This represents the Hadamard product operation; express The hidden state at time T. The hidden state output at the last time step T is used. This is the vector representing the entire packet.

[0044] Step 3: Determine the first fused representation vector and the second fused representation vector based on the potential space, the cell representation vector, and the whole package representation vector.

[0045] Among them, the first fused representation vector is the cell representation vector after fused whole package representation vector; the second fused representation vector is the whole package representation vector after fused individual cell representation vectors.

[0046] Specifically, an attention mechanism is used within the latent space to fuse the cell representation vector and the entire package representation vector, resulting in a first fused representation vector and a second fused representation vector. For example, an attention mechanism is used within the latent space to fuse the representation vector of a single cell branch. and the representation vector of the entire branch To integrate. Among them, , The fused vector is then decoded to restore the original temperature data, and the residual between the restored temperature data and the original temperature data is calculated, as follows: First, based on the linear transformation weights (These linear transformation weights are all) The query (query, q), key (key, k), and value (value, v) matrices are generated as follows: ; ; in, It is a positive integer.

[0047] Then, the representation vectors of each cell are fused onto the overall package representation vector. The specific process is as follows: ; ; ; in, This represents a vector concatenation operation; This represents the entire package representation vector after fusing the representation vectors of each cell.

[0048] The entire package representation vector is fused from the individual cell representation vectors. The specific process is as follows: ; ; ; in, This represents the cell representation vector after the fusion of the entire package representation vector.

[0049] Step 4: Determine the first reconstruction residual and the second reconstruction residual based on the first fusion representation vector, the second fusion representation vector, the decoder, historical temperature data, and historical temperature statistics data.

[0050] The decoder includes a first decoder branch and a second decoder branch. The first fusion represents a vector input to the first decoder branch, and the second fusion represents a vector input to the second decoder branch. The decoder's structure is the inverse of the encoder's structure.

[0051] In some embodiments, determining a first reconstruction residual and a second reconstruction residual based on a first fused representation vector, a second fused representation vector, a decoder, historical temperature data, and historical temperature statistics data includes: inputting the first fused representation vector and the second fused representation vector into a decoder to obtain cell temperature reconstruction data and whole-pack statistics reconstruction data; determining the first reconstruction residual based on the cell temperature reconstruction data and historical temperature data; and determining the second reconstruction residual based on the whole-pack statistics reconstruction data and historical temperature statistics data.

[0052] Specifically, the data is input into the corresponding decoders to obtain the reconstruction results. The reconstruction results include reconstructed cell temperature data and reconstructed package statistics data. Then, the residuals between the reconstructed temperature data and the original temperature data are calculated; that is, the first reconstruction residual is calculated between the reconstructed cell temperature data and the historical temperature data, and the second reconstruction residual is calculated between the reconstructed package statistics data and the historical temperature statistics data. The specific calculation formulas are as follows: ; ; in, Indicates a single battery cell exist The original value of the temperature at that moment; Indicates a single battery cell exist Reconstructed temperature value at any given time; Indicates a single battery cell The reconstructed residuals; Indicates the whole package is The original values ​​of the time-time statistics vector; Indicates the whole package is The reconstructed value of the time-time statistics vector; This represents the reconstruction residual of the entire package.

[0053] Step 5: Using a preset loss function, train the first and second reconstruction residuals to obtain the target detection model.

[0054] The formula for calculating the preset loss function is as follows: ; in, Indicates the preset loss function; and This represents the allocation coefficient between cell loss and overall package loss. =1; This represents the number of statistical features in the entire packet vector; This represents the L2 regularization coefficient, which prevents overfitting during training. This represents the set of network parameters.

[0055] It should be noted that the above-mentioned package temperature statistics data in this application embodiment can also include other variables that can reflect the temperature state of the package, such as other mathematical statistics, such as the median, 25th quantile, 75th quantile, etc. These can be set according to actual conditions and are not specifically limited here. The autoencoder structure described in this application embodiment can also be other neural network structures suitable for time-series data, such as Long Short-Term Memory (LSTM) networks, Transformer models, etc. These can be set according to actual conditions and are not specifically limited here.

[0056] Step 120: Determine the first target abnormal threshold for the battery cell and the second target abnormal threshold for the battery pack based on the preset operating condition correction strategy, the first preset abnormal threshold and the second preset abnormal threshold.

[0057] The first and second preset anomaly thresholds are determined during the target detection model training phase, as follows: In some embodiments, the battery pack temperature state detection method further includes: determining a first preset anomaly threshold based on the mean and standard deviation of all first reconstruction residuals of the battery cells within a first preset time period; and determining a second preset anomaly threshold based on the mean and standard deviation of all second reconstruction residuals of the battery pack within a first preset time period.

[0058] The first preset duration can be the historical duration corresponding to the historical temperature data of the battery cell (or the historical duration corresponding to the historical temperature statistics data of the battery pack), or other values. The specific settings can be made according to the actual situation, and no specific limitations are made here.

[0059] Specifically, based on the residual distribution of the training set, 3 The rules determine the thresholds (including a first preset anomaly threshold and a second preset anomaly threshold, both determined using the same method), as follows: For the set of residuals of a single battery cell (i.e., all the first reconstruction residuals of the battery cell within the first preset time period) And the entire set of cell residuals (i.e., all second reconstruction residuals of the battery pack within the first preset time period). Each entity calculates its corresponding anomaly detection threshold using the following formula: ; ; in, and Indicates the first The mean and standard deviation of all residuals for each cell; Indicates the first The threshold for determining anomalies in individual battery cells; and This represents the mean and standard deviation of all residuals in the entire package; This represents the threshold for determining anomalies in the entire package.

[0060] In some embodiments, determining a first target abnormal threshold for the battery cell and a second target abnormal threshold for the battery pack based on a preset operating condition correction strategy, a first preset abnormal threshold, and a second preset abnormal threshold includes: determining an operating condition correction factor based on the current operating condition of the battery pack; the current operating condition being either an extreme or non-extreme operating condition; determining the first target abnormal threshold based on the operating condition correction factor and the first preset abnormal threshold; and determining the second target abnormal threshold based on the operating condition correction factor and the second preset abnormal threshold.

[0061] The preset operating condition correction strategy refers to determining the operating condition correction factor based on the current operating condition of the battery pack. For example, in extreme operating condition scenarios, the operating condition correction factor is used to amplify the thresholds (including the first preset abnormal threshold and the second preset abnormal threshold). Extreme operating condition scenarios include high-rate current charging and discharging and extreme ambient temperatures. For instance, in high-temperature environments (ambient temperature above 50°C), the battery temperature rise rate is significantly higher than in normal temperature environments. Using the original threshold for judgment in this situation may lead to misjudgments. The operating condition correction factor can amplify the abnormal judgment threshold to achieve threshold correction in high-temperature environments. The specific calculation formula is as follows: ; in, This represents the operating condition correction factor, which is usually set to 1.1. Under other extreme operating conditions, it can be set according to the actual situation, but the maximum value should not exceed 1.25. This represents the corrected abnormal threshold for a single battery cell, i.e., the first target abnormal threshold. This represents the corrected overall package anomaly threshold, i.e., the second target anomaly threshold.

[0062] Step 130: If the first current residual is greater than the first target anomaly threshold, determine that the cell temperature is abnormal; if the second current residual is greater than the second target anomaly threshold, determine that the battery pack temperature is abnormal.

[0063] Specifically, by inputting the current temperature data of each battery cell into the trained target detection model, the first current residual of each cell can be obtained. Furthermore, based on the preset operating condition correction strategy and the first preset anomaly threshold for each cell, the first target anomaly threshold for each cell can be obtained. Finally, the first current residual of each cell is compared with the corresponding first target anomaly threshold. If the first current residual is greater than the corresponding first target anomaly threshold, it indicates that the temperature state of that cell is abnormal; otherwise, it indicates that the temperature state of that cell is normal. This allows for accurate detection of the temperature state of each battery cell. Similarly, by inputting the current temperature statistics of the battery pack into the trained target detection model, the second current residual of the battery pack can be obtained. Furthermore, based on the preset operating condition correction strategy and the second preset anomaly threshold of the battery pack, the second target anomaly threshold of the battery pack can be obtained. Finally, the second current residual of the battery pack is compared with the corresponding second target anomaly threshold. If the second current residual is greater than the corresponding second target anomaly threshold, it indicates that the temperature state of the battery pack is abnormal; otherwise, it indicates that the temperature state of the battery pack is normal. This enables accurate detection of the temperature state of the battery pack, thereby achieving multi-scale temperature anomaly detection at the battery pack and individual cell levels.

[0064] It is understood that the battery pack temperature state detection method provided in this application uses a target detection model trained based on an autoencoder and normal state data of the battery pack to output a first current residual of the battery cell and a second current residual of the battery pack. The method determines whether the temperature of the battery cell is abnormal by comparing the first current residual with a first target abnormality threshold, and determines whether the temperature of the battery pack is abnormal by comparing the second current residual with a second target abnormality threshold. This achieves accurate detection of the temperature state of the battery pack and each battery cell.

[0065] It should be noted that the battery pack temperature detection method in this application embodiment also includes the introduction of a continuous window consistency check, that is, an early warning is only issued if the threshold is exceeded within a specified window length. The time window length can be set according to the actual data sampling interval and sample length.

[0066] In some embodiments, the battery pack temperature detection method further includes: determining a first residual mean and a first residual standard deviation based on a first current residual of the battery cell at each time step within a second preset time period; determining a second residual mean and a second residual standard deviation based on a second current residual of the battery pack at each time step within a second preset time period; and issuing a warning or graded alarm when the first residual mean and the first residual standard deviation, and / or the second residual mean and the second residual standard deviation, meet preset alarm conditions.

[0067] The second preset duration is the duration of a continuous time window (including multiple continuous time steps). The specific value can be set according to the actual situation, and no specific limit is made here.

[0068] Specifically, the mean and standard deviation of the residuals within a certain historical time window at the current time step are calculated. This allows for the acquisition of a residual mean and standard deviation at each time step (for example, determining the first residual mean and standard deviation based on the first current residual of the battery cell at each time step within a second preset time period; determining the second residual mean and standard deviation based on the second current residual of the battery pack at each time step within a second preset time period). The residual mean directly measures the average deviation of the temperature curve from the normal mode over a recent period, thus corresponding to the severity of the anomaly. The standard deviation reflects the volatility of the deviation over the same period; its rise or fall can reveal whether the fault is still rapidly evolving or has stabilized. Therefore, the residual mean... As an indicator of anomaly severity (for example, using the mean of the first residuals of each cell as the anomaly severity indicator for the corresponding cell, and the mean of the second residuals of the battery pack as the anomaly severity indicator for the battery pack), the standard deviation of the residuals is used. The changing trend serves as an indication of uncertainty (for example, the first residual standard deviation of each cell is used as an indication of uncertainty for that cell, and the second residual standard deviation of the battery pack is used as an indication of uncertainty for the entire battery pack), enabling early warning and tiered alarms for temperature anomalies. A larger residual mean indicates a greater deviation of the overall temperature curve from the normal pattern. A higher residual standard deviation indicates greater fluctuations in the residual value within a continuous window, while a lower residual standard deviation indicates greater residual stability.

[0069] Figure 3 This is a schematic diagram illustrating the early warning and tiered warning processes provided in the embodiments of this application. For example, see [link to relevant documentation]. Figure 3 Taking the judgment of early warning or graded alarm for a battery cell as an example, the process of determining whether an early warning or graded alarm is needed for the battery cell based on the mean of the first residual, the standard deviation of the first residual, and the preset alarm conditions is as follows: when the mean of the first residual is close to the first target abnormal threshold (0.8-1), the alarm is triggered by the following conditions: If the mean of the first residual value is greater than the first target anomaly threshold and the standard deviation of the first residual value continues to rise, then an early warning for the battery cell is required, i.e., the warning stage begins. If the mean of the first residual value is greater than the first target anomaly threshold and the standard deviation of the first residual value continues to rise, then a Level 1 alarm for the battery cell is required. If the mean of the first residual value is greater than the first preset multiple (e.g., 1.5 times) of the first target anomaly threshold and the standard deviation of the first residual value shows a downward trend or a stable range, then a Level 2 alarm for the battery cell is required. If the mean of the first residual value is greater than the second preset multiple (e.g., 1.5 times) of the first target anomaly threshold and the standard deviation of the first residual value shows an upward trend, then a Level 3 alarm for the battery cell is required. If the mean of the first residual value is greater than the third preset multiple (e.g., 2 times) of the first target anomaly threshold and the standard deviation of the first residual value remains unchanged, then a Level 4 alarm for the battery cell is required. Similarly, the principle for judging early warnings or graded alarms for battery packs is the same and will not be elaborated further here.

[0070] Figure 4 This is a schematic flowchart illustrating the overall process of a battery pack temperature state detection method provided in an embodiment of this application. For example, see [link to relevant documentation]. Figure 4 The overall process of this battery pack temperature state detection method includes: First, collecting cell temperatures, overall pack temperature statistics, and operating condition information under various operating scenarios, and constructing a dataset containing only normal conditions by slicing the data into fixed time windows. Then, using the individual cell temperature sequence and operating condition information as input, a single-cell autoencoder branch with shared parameters is established to learn the normal evolution pattern of each cell. Using the overall pack temperature statistics and operating condition information as input, an overall pack autoencoder branch is established to learn the normal pattern of the overall thermal characteristics. Second, the representations of the single-cell branch and the overall pack branch are fused in the latent space, and the residuals between the temperature data in the two types of reconstructed outputs and the original temperature data are calculated. Thresholds are automatically generated based on the statistical characteristics of the residuals, and a continuous window consistency check and operating condition correction strategy are introduced. Finally, the mean residual is used as an indicator of anomaly severity, and the trend of the standard deviation of the residual is used as an uncertainty indicator to achieve early warning and graded alarm for temperature anomalies.

[0071] In summary, the battery pack temperature state detection method provided in this application is trained using only normal samples, allowing the target detection model to learn the thermal behavior patterns of the entire pack and individual cells under normal conditions. It replaces fixed thresholds with residual statistical thresholds, fundamentally eliminating reliance on fault labels. Furthermore, by training two types of autoencoders using temperature data from the battery pack and individual cells under normal conditions, the individual cell branch and the entire pack branch learn their respective temperature evolution patterns. Therefore, the two types of residuals tend to be concentrated under normal conditions, with relatively small variance. When any unlearned thermal behavior occurs, such as a rapid temperature rise in individual cells due to a local short circuit, or an overall temperature rise caused by heat dissipation failure, the target detection model will be unable to accurately reconstruct the data, and the corresponding residuals will be amplified, thus ensuring high sensitivity.

[0072] Furthermore, by employing bidirectional attention in the latent space, the overall package characterization can dynamically monitor cells with significant temperature fluctuations, while simultaneously incorporating the overall temperature changes into the characterization of each cell. This cyclical reinforcement mechanism, combining local and global approaches, prevents the overall mean from diluting local anomalies and avoids local noise misleading the overall judgment, thereby simultaneously reducing both false positive and false negative rates. Additionally, the threshold used to determine whether a temperature state is abnormal is adaptively determined based on the statistical characteristics of the residuals, supplemented by continuous window consistency checks and operating condition correction strategies. This allows the judgment rules to automatically adjust according to actual operating conditions, maintaining adaptability to extreme operating conditions while retaining high sensitivity to genuine anomalies. Therefore, through the reconstruction of the statistical boundaries of the residuals, the information complementarity of contextual attention, and the adaptive calibration of the threshold, the method of this application significantly outperforms related technologies that employ fixed thresholds and supervised learning methods dependent on fault labels in terms of sensitivity, stability, and robustness.

[0073] In summary, the key technical points of the battery pack temperature state detection method provided in this application embodiment are: First, this application embodiment only uses data from fault-free samples for training, thus eliminating the dependence on fault labels.

[0074] Secondly, in this application embodiment, two types of autoencoders are used to reconstruct the temperature sequence of individual cells and the statistical sequence of the whole package, respectively. The two are independent of each other but intersect and merge in the potential space. The whole package representation vector dynamically focuses on cells with large temperature fluctuations, while the cell representation vector synchronously absorbs the overall temperature changes, avoiding the mean from masking local anomalies and suppressing noise misjudgment.

[0075] Third, the embodiments of this application provide a threshold based on the statistical distribution of the reconstruction error, which eliminates the need for manual calibration. Furthermore, the continuous window consistency check and operating condition correction strategy are adopted to ensure that the method works in a balanced manner under all operating conditions.

[0076] Figure 5This is a schematic block diagram of a battery pack temperature state detection system provided in an embodiment of this application. This application also provides a battery pack temperature state detection system, see below. Figure 5 The battery pack temperature status detection system 100 includes: a first determining module 101, used to determine a first current residual of the battery cell and a second current residual of the battery pack based on the current temperature data of the battery cell, the current temperature statistics data of the battery pack, and a target detection model; wherein the target detection model is trained based on an autoencoder and normal state data of the battery pack; a second determining module 102, used to determine a first target abnormality threshold of the battery cell and a second target abnormality threshold of the battery pack based on a preset operating condition correction strategy, a first preset abnormality threshold, and a second preset abnormality threshold; a third determining module 103, used to determine that the battery cell temperature is abnormal when the first current residual is greater than the first target abnormality threshold; and a fourth determining module 104, used to determine that the battery pack temperature is abnormal when the second current residual is greater than the second target abnormality threshold.

[0077] The technical solution of this application provides a battery pack temperature status detection system. By using a target detection model trained based on an autoencoder and normal state data of the battery pack, the system outputs a first current residual of the battery cell and a second current residual of the battery pack. The system determines whether the temperature of the battery cell is abnormal by comparing the first current residual with a first target abnormality threshold, and determines whether the temperature of the battery pack is abnormal by comparing the second current residual with the second target abnormality threshold. This enables accurate detection of the temperature status of the battery pack and each battery cell.

[0078] In some embodiments, normal state data includes historical temperature data of the battery cell, historical temperature statistics data of the battery pack, and operating condition information of the battery pack; the autoencoder includes a first autoencoder branch, a second autoencoder branch, a latent space, and a decoder; the battery pack temperature state detection system 100 further includes a fifth determining module, used for: establishing a first autoencoder branch for the battery cell and a second autoencoder branch for the battery pack; determining the cell representation vector of the battery cell and the whole-pack representation vector of the battery pack based on historical temperature data, historical temperature statistics data, operating condition information, the first autoencoder branch, and the second autoencoder branch; determining a first fused representation vector and a second fused representation vector based on the latent space, the cell representation vector, and the whole-pack representation vector; wherein, the first fused representation vector is the cell representation vector after fusing the whole-pack representation vector; the second fused representation vector is the whole-pack representation vector after fusing the cell representation vectors; determining a first reconstruction residual and a second reconstruction residual based on the first fused representation vector, the second fused representation vector, the decoder, historical temperature data, and historical temperature statistics data; and training the first reconstruction residual and the second reconstruction residual using a preset loss function to obtain a target detection model.

[0079] In some embodiments, the first autoencoder branch includes a temporal convolutional network; the second autoencoder branch includes a gated loop unit; the fifth determining module is further configured to: determine the cell representation vector based on historical temperature data, operating condition information and the temporal convolutional network; and determine the whole package representation vector based on historical temperature statistics data, operating condition information and the gated loop unit.

[0080] In some embodiments, the fifth determining module is further configured to: input the first fused representation vector and the second fused representation vector into the decoder to obtain cell temperature reconstruction data and whole package statistical reconstruction data; determine the first reconstruction residual based on the cell temperature reconstruction data and historical temperature data; and determine the second reconstruction residual based on the whole package statistical reconstruction data and historical temperature statistical data.

[0081] In some embodiments, the battery pack temperature state detection system is further configured to: determine a first preset anomaly threshold based on the mean and standard deviation of all first reconstruction residuals of the battery cells within a first preset time period; and determine a second preset anomaly threshold based on the mean and standard deviation of all second reconstruction residuals of the battery pack within a first preset time period.

[0082] In some embodiments, the battery pack temperature status detection system is further configured to: determine a first residual mean and a first residual standard deviation based on the first current residual of the cell at each time step within a second preset time period; determine a second residual mean and a second residual standard deviation based on the second current residual of the battery pack at each time step within a second preset time period; and, if the first residual mean and the first residual standard deviation, and / or the second residual mean and the second residual standard deviation, satisfy a preset alarm condition, issue a warning or a graded alarm.

[0083] In some embodiments, the second determining module is further configured to: determine a condition correction factor based on the current operating condition of the battery pack; the current operating condition is an extreme operating condition or a non-extreme operating condition; determine a first target abnormal threshold based on the condition correction factor and a first preset abnormal threshold; and determine a second target abnormal threshold based on the condition correction factor and a second preset abnormal threshold.

[0084] This embodiment also provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, it implements the method of any of the above embodiments.

[0085] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps of any of the methods in the above embodiments.

[0086] In the embodiments of this application, the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0087] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0088] The above provides a detailed description of a battery pack temperature state detection method, system, device, and storage medium provided in the embodiments of 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 core ideas of this application. 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 detecting the temperature state of a battery pack, characterized in that, The battery pack includes multiple battery cells; the method includes: Based on the current temperature data of the battery cell, the current temperature statistics of the battery pack, and the target detection model, the first current residual of the battery cell and the second current residual of the battery pack are determined; wherein, the target detection model is trained based on the autoencoder and the normal state data of the battery pack; Based on the preset operating condition correction strategy, the first preset abnormal threshold, and the second preset abnormal threshold, the first target abnormal threshold of the battery cell and the second target abnormal threshold of the battery pack are determined. If the first current residual is greater than the first target anomaly threshold, the cell temperature is determined to be abnormal. If the second current residual is greater than the second target anomaly threshold, the battery pack temperature is determined to be abnormal.

2. The method according to claim 1, characterized in that, The normal state data includes the historical temperature data of the battery cell, the historical temperature statistics of the battery pack, and the operating condition information of the battery pack; the self-encoder includes a first self-encoder branch, a second self-encoder branch, a latent space, and a decoder; The method for determining the target detection model includes: Establish the first self-encoder branch of the battery cell and the second self-encoder branch of the battery pack; Based on the historical temperature data, the historical temperature statistics data, the operating condition information, the first autoencoder branch, and the second autoencoder branch, the cell representation vector of the battery cell and the whole pack representation vector of the battery pack are determined. Based on the potential space, the cell representation vector, and the package representation vector, a first fused representation vector and a second fused representation vector are determined; wherein, the first fused representation vector is the cell representation vector after fusing the package representation vector; and the second fused representation vector is the package representation vector after fusing each of the cell representation vectors. Based on the first fusion representation vector, the second fusion representation vector, the decoder, the historical temperature data, and the historical temperature statistics data, determine the first reconstruction residual and the second reconstruction residual; The target detection model is obtained by training the first reconstruction residual and the second reconstruction residual using a preset loss function.

3. The method according to claim 2, characterized in that, The first autoencoder branch includes a temporal convolutional network; the second autoencoder branch includes a gated recurrent unit; determining the cell representation vector of the battery cell and the overall battery pack representation vector based on the historical temperature data, the historical temperature statistics data, the operating condition information, the first autoencoder branch, and the second autoencoder branch includes: The cell representation vector is determined based on the historical temperature data, the operating condition information, and the temporal convolutional network. The whole package representation vector is determined based on the historical temperature statistics, the operating condition information, and the gated loop unit.

4. The method according to claim 2, characterized in that, The step of determining the first reconstruction residual and the second reconstruction residual based on the first fusion representation vector, the second fusion representation vector, the decoder, the historical temperature data, and the historical temperature statistics data includes: The first fused representation vector and the second fused representation vector are input into the decoder to obtain cell temperature reconstruction data and whole pack statistics reconstruction data; The first reconstruction residual is determined based on the cell temperature reconstruction data and the historical temperature data; The second reconstruction residual is determined based on the reconstructed data of the whole package statistics and the historical temperature statistics data.

5. The method according to claim 2, characterized in that, The method further includes: determining the first preset anomaly threshold based on the mean and standard deviation of all the first reconstruction residuals of the battery cell within a first preset time period; The second preset anomaly threshold is determined based on the mean and standard deviation of all the second reconstruction residuals of the battery pack within the first preset time period.

6. The method according to claim 1, characterized in that, The method further includes: determining the first residual mean and the first residual standard deviation based on the first current residual of the cell at each time step within a second preset time period; Based on the second current residual of the battery pack at each time step within the second preset time period, determine the mean of the second residual and the standard deviation of the second residual; If the mean of the first residual and the standard deviation of the first residual, and / or the mean of the second residual and the standard deviation of the second residual, meet the preset alarm conditions, an early warning or a graded alarm will be issued.

7. The method according to claim 1, characterized in that, The step of determining the first target anomaly threshold of the battery cell and the second target anomaly threshold of the battery pack based on a preset operating condition correction strategy, a first preset anomaly threshold, and a second preset anomaly threshold includes: The operating condition correction factor is determined based on the current operating condition of the battery pack; the current operating condition is either an extreme operating condition or a non-extreme operating condition. The first target anomaly threshold is determined based on the operating condition correction factor and the first preset anomaly threshold. The second target anomaly threshold is determined based on the operating condition correction factor and the second preset anomaly threshold.

8. A battery pack temperature status detection system, characterized in that, The battery pack includes multiple battery cells, and the system includes: The first determining module is used to determine the first current residual of the battery cell and the second current residual of the battery pack based on the current temperature data of the battery cell, the current temperature statistics data of the battery pack, and the target detection model; wherein, the target detection model is trained based on the autoencoder and the normal state data of the battery pack; The second determining module is used to determine the first target abnormal threshold of the battery cell and the second target abnormal threshold of the battery pack based on the preset working condition correction strategy, the first preset abnormal threshold and the second preset abnormal threshold. The third determining module is used to determine that the cell temperature is abnormal when the first current residual is greater than the first target abnormal threshold. The fourth determining module is used to determine that the battery pack temperature is abnormal when the second current residual is greater than the second target abnormality threshold.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to perform the steps of the method as described in any one of claims 1-7.

Images