A battery temperature management method, apparatus, device and medium

Abstract

This application discloses a battery temperature management method, apparatus, device, and medium, comprising: acquiring battery state parameters and encoding the state parameters to generate a word sequence, wherein the state parameters include the monitored temperatures of various monitoring parts of the battery; processing the word sequence using a directed thermal conduction graph of the battery to construct a textual sequence; inputting the textual sequence into a prediction model to obtain the predicted temperatures of various monitoring parts of the battery; and providing a temperature warning for the battery based on the monitored temperatures and predicted temperatures. This application integrates the physical thermal conduction mechanism into the input data of the prediction model, enabling the prediction model to understand the generation, propagation, and dissipation of heat, enhancing the prediction model's ability to learn early signals of heat diffusion, thereby achieving early and accurate over-temperature warnings.

Description

Technical Field

[0001] This application relates to the field of battery management technology, and in particular to a battery temperature management method, apparatus, device, and medium. Background Technology

[0002] With the increasing global demand for clean energy, electric vehicles, hybrid vehicles, and various energy storage systems that rely on power batteries as their core energy source have been widely adopted.

[0003] During charging and discharging, the temperature of the power battery gradually rises due to internal electrochemical reactions and the heat generated by current flowing through internal resistance. When the temperature exceeds the normal operating range, it can trigger a series of serious problems. On the one hand, high temperatures accelerate the aging and decomposition of internal battery materials, leading to irreversible capacity decay and shortening the overall battery lifespan. On the other hand, excessively high temperatures may trigger thermal runaway reactions within the battery, causing it to bulge, rupture, or even explode, posing a significant threat to personal safety and the safety of equipment and property.

[0004] It is evident that early and accurate over-temperature warning for batteries is an urgent problem to be solved. Summary of the Invention

[0005] This application provides a battery temperature management method, apparatus, device, and medium that can provide early and accurate over-temperature warning for batteries.

[0006] In a first aspect, embodiments of this application provide a battery temperature management method, including: The battery status parameters are acquired and encoded to generate a word sequence. The status parameters include the monitoring temperature of each monitoring part of the battery. The word sequence is processed using the directed graph of thermal conduction of the battery to construct a textual sequence; The textual sequence is input into the prediction model to obtain the predicted temperature of each monitoring part of the battery; Based on the monitored and predicted temperatures, a temperature warning is issued for the battery.

[0007] In conjunction with the first aspect, in one implementation, obtaining the battery's state parameters includes: Collect the operating parameters of the battery; The current operating condition type of the battery is identified based on the operating parameters; The data acquisition strategy is dynamically adjusted based on the current operating condition type. The state parameters are collected based on the adjusted data acquisition strategy.

[0008] In conjunction with the first aspect, in one implementation, the current operating condition type includes at least two of the following: a stationary operating condition, a momentary operating condition, and a steady-state driving condition; Dynamically adjust the data acquisition strategy according to the current operating condition type, including: The sampling frequency is dynamically adjusted in response to changes in the current operating condition type. Specifically, when the vehicle is stationary, the sampling frequency is reduced; when the vehicle is in steady-state driving condition, the normal sampling frequency is maintained; and when the vehicle is in transient condition, the sampling frequency is increased.

[0009] In conjunction with the first aspect, in one embodiment, the operating parameters include charging and discharging current; identifying the current operating condition type of the battery based on the operating parameters includes: If the absolute value of the charging and discharging current is less than the preset static charging and discharging current threshold and the duration is greater than the preset static time, it is determined to be a static working condition. If the rate of change of the charging and discharging current is greater than the preset rapid acceleration rate of change, it is determined to be a rapid acceleration condition under instantaneous operating conditions. If the rate of change of the charging and discharging current is less than the preset rate of change of rapid deceleration, it is determined to be a rapid deceleration condition under instantaneous operating conditions. If the charging and discharging current is less than 0 and the absolute value of the charging and discharging current is greater than the preset charging and discharging current threshold, it is determined to be a charging condition under instantaneous operating conditions. If neither of the above conditions meets the requirements for static or instantaneous operation, the condition is determined to be a steady-state driving condition.

[0010] In conjunction with the first aspect, in one implementation, the current operating condition type includes at least two of the following: a stationary operating condition, a momentary operating condition, and a steady-state driving condition; Encoding the state parameters to generate a word sequence includes: When in a static operating condition, the state parameters within the statistical time window are statistically compressed and encoded to generate lexical units that represent statistical features, and the lexical units are arranged in the order of generation to form a lexical unit sequence. When in steady-state or instantaneous driving conditions, the state parameters of each sampling point are mapped and encoded to generate lexical units that represent instantaneous features. The lexical units are then arranged in the order of generation to form a lexical unit sequence.

[0011] In conjunction with the first aspect, in one embodiment, the method further includes: Based on the physical structure of the battery, a directed heat conduction graph with directional nodes and directed edges is constructed; The physical structure includes the monitoring part, the thermal resistance network inside the battery, and the cooling structure.

[0012] In conjunction with the first aspect, in one embodiment, the term sequence is processed using the directed graph of the battery's thermal conduction to construct a textual sequence, including: A temperature flow sequence is constructed based on the lexical units corresponding to the monitored temperature and the directed graph of heat conduction, and the temperature flow sequence is used as the textual sequence.

[0013] In conjunction with the first aspect, in one embodiment, the state parameters further include the battery's operating parameters. The lexical sequence is processed using the directed graph of the battery's thermal conduction to construct a textualized sequence, including: A temperature flow sequence is constructed based on the terms corresponding to the monitored temperature and the directed graph of heat conduction; The temperature flow sequence is concatenated with the word sequence corresponding to the operating parameters to form the textual sequence.

[0014] In conjunction with the first aspect, in one implementation, constructing a temperature flow sequence based on the lexical units corresponding to the monitored temperature and the directed heat conduction graph includes: For each directed edge in the heat conduction directed graph, calculate the theoretical delay time; For each target time, the historical time corresponding to the heat source node on the directed edge is determined based on the theoretical delay time; Arrange the tokens corresponding to the monitored temperature of the heat source node at the historical time and the tokens corresponding to the monitored temperature of the heated node on the directed edge at the target time to form a causal sequence block; The temperature flow sequence is formed by splicing together multiple causal sequence blocks in the time order of the target time.

[0015] In conjunction with the first aspect, in one implementation, arranging the tokens corresponding to the monitored temperature of the heat source node at the historical moment and the tokens corresponding to the monitored temperature of the heated node on the directed edge at the target moment to form a causal sequence block includes: Calculate the integer number of time steps corresponding to the theoretical delay time; Take the nearest time step of the integer time step, and arrange the words corresponding to the monitored temperature of the heat source node at the historical moment corresponding to the nearest time step in sequence before the words corresponding to the monitored temperature of the heated node at the target moment to form the causal sequence block.

[0016] In conjunction with the first aspect, in one implementation, the prediction model includes an attention mechanism in which a spatial mask is embedded, comprising: Construct an adjacency matrix where nodes with direct directed edges are assigned a value of 1, and nodes without direct directed edges are assigned a value of 0. When calculating attention weights, a preset mask value is added to the scores of nodes that do not have direct directed edges, so that the attention weights of the nodes that do not have direct directed edges are zero.

[0017] In conjunction with the first aspect, in one implementation, the textualized sequence is input into a prediction model to obtain the predicted temperature of each monitoring component of the battery, including: The embedding layer of the prediction model maps the lexical units in the textual sequence into a high-dimensional feature representation of a unified dimension. A prediction input window is constructed based on a preset number of high-dimensional feature representations. The window slides on the textual sequence with a sliding step size to make predictions, thereby obtaining the predicted temperature of each monitoring part of the battery in the future time window.

[0018] In conjunction with the first aspect, in one implementation, a temperature warning is provided for the battery based on the monitored temperature and the predicted temperature, including: The high-dimensional feature representation corresponding to the end of the prediction input window is selected and compared with the predicted temperature at the end of the future time window to calculate the temperature deviation. The warning level is determined based on the temperature deviation. The corresponding warning will be issued based on the warning level.

[0019] Secondly, embodiments of this application provide a battery temperature management device, comprising: The parameter acquisition module is used to acquire the state parameters of the battery and encode the state parameters to generate a word sequence. The state parameters include the monitoring temperature of each monitoring part of the battery. A text generation module is used to process the word sequence using the directed graph of thermal conduction of the battery to construct a textual sequence; A temperature prediction module is used to input the textual sequence into the prediction model to obtain the predicted temperature of each monitoring part of the battery. An early warning module is used to provide temperature warnings for the battery based on the monitored temperature and the predicted temperature.

[0020] Thirdly, embodiments of this application provide a battery temperature management device, the battery temperature management device including a processor, a memory, and a battery temperature management program stored in the memory and executable by the processor, wherein when the battery temperature management program is executed by the processor, it implements the steps of the battery temperature management method as described in any of the preceding claims.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium storing a battery temperature management program, wherein when the battery temperature management program is executed by a processor, it implements the steps of the battery temperature management method as described in any of the preceding claims.

[0022] The beneficial effects of the technical solution provided in this application include: This application provides a battery temperature management method, apparatus, device, and medium. In this embodiment, battery state parameters are encoded to generate a word sequence, and a textual sequence with causal relationships in heat conduction is constructed using the battery's directed thermal conduction graph. This integrates the physical heat conduction mechanism into the input data of the prediction model, enabling the model to understand the generation, propagation, and dissipation of heat. This solves the problem of large prediction errors in heat propagation paths caused by traditional models ignoring the physical topology of battery heat conduction, and enhances the prediction model's ability to learn early signals of heat diffusion. Based on this, the textual sequence is input into the prediction model to obtain the predicted temperature of each monitoring location. Temperature warnings are then issued based on the deviation between the actual monitored temperature and the predicted temperature, overcoming the shortcomings of traditional fixed-threshold warning methods, such as high lag and inability to identify potential thermal risks in advance. This effectively reduces the false alarm rate and the missed alarm rate. This embodiment achieves advanced and accurate early warnings that conform to physical laws, significantly improving the accuracy, timeliness, and safety of battery temperature management, providing reliable technical support for preventing battery thermal runaway, and ensuring the stable operation of the battery system under complex operating conditions. Attached Figure Description

[0023] 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.

[0024] Figure 1 This is a flowchart illustrating an embodiment of the battery temperature management method of this application; Figure 2 This is a schematic diagram visualizing the directed graph of heat conduction in this application; Figure 3 This is a functional module diagram of an embodiment of the battery temperature management device of this application; Figure 4 This is a schematic diagram of the hardware structure of the battery temperature management device involved in the embodiments of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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, 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.

[0026] In a first aspect, embodiments of this application provide a battery temperature management method.

[0027] In one embodiment, reference is made to Figure 1 , Figure 1 This is a schematic flowchart of an embodiment of the battery temperature management method of this application. Figure 1 As shown, battery temperature management methods include: S101: Obtain the state parameters of the battery and encode the state parameters to generate a word sequence. The state parameters include the monitoring temperature of each monitoring part of the battery.

[0028] In step S101, the battery being monitored can be a power battery in an electric vehicle, another battery in an electric vehicle that needs to be monitored for temperature, or a battery in other equipment that needs to be monitored for temperature. This application does not specifically limit the type of battery.

[0029] Based on the battery's structural characteristics and thermal conductivity, temperature sensors are distributed across the battery.

[0030] The sensors include contact sensors (such as T-type thermocouples and platinum resistance thermometers) and non-contact sensors (such as infrared temperature sensors). In the battery module, contact sensors are installed on the positive and negative terminals, sides, and connection points between modules of each cell; non-contact sensors are installed on the top and bottom of the battery pack casing and near the heat sink fins. Contact sensors can accurately measure the temperature at the contact points, while non-contact sensors are used to detect the overall temperature distribution and heat dissipation effect.

[0031] S102: The word sequence is processed using the directed graph of thermal conduction of the battery to construct a textual sequence.

[0032] Since the directed graph of heat conduction has a causal relationship of heat conduction, the constructed textual sequence also has a causal relationship of heat conduction.

[0033] S103: Input the textual sequence into the prediction model to obtain the predicted temperature of each monitoring part of the battery.

[0034] S104: Based on the monitored temperature and predicted temperature, provide a temperature warning for the battery.

[0035] In this embodiment, a word sequence is generated by encoding battery state parameters, and a textual sequence with causal relationships in heat conduction is constructed using the directed graph of battery thermal conduction. This integrates the physical heat conduction mechanism into the input data of the prediction model, enabling the model to understand the generation, propagation, and dissipation of heat. This solves the problem of large prediction errors in heat propagation paths caused by traditional models ignoring the physical topology of battery heat conduction, and enhances the model's ability to learn early signals of heat diffusion. Based on this, the textual sequence is input into the prediction model to obtain the predicted temperature of each monitoring location. Temperature warnings are then issued based on the deviation between the actual monitored temperature and the predicted temperature, overcoming the shortcomings of traditional fixed-threshold warning methods, such as high lag and inability to identify potential thermal risks in advance. This effectively reduces the false alarm rate and the missed alarm rate. This embodiment achieves advanced and accurate early warnings that conform to physical laws, significantly improving the accuracy, timeliness, and safety of battery temperature management, providing reliable technical support for preventing battery thermal runaway, and ensuring the stable operation of the battery system under complex operating conditions.

[0036] Furthermore, in one embodiment, obtaining the battery's state parameters includes: S201: Collect the operating parameters of the battery.

[0037] S202: Identify the current operating condition type of the battery based on the operating parameters.

[0038] S203: Dynamically adjust the data acquisition strategy according to the current operating condition type.

[0039] S204: Collect the state parameters based on the adjusted data acquisition strategy.

[0040] In this embodiment, the current operating condition type is identified by collecting operating parameters, and the data acquisition strategy is dynamically adjusted accordingly. This solves the problem of resource waste or missing key data caused by the inability of traditional fixed sampling strategies to adapt to the changing battery operating conditions. It achieves adaptive switching between low-frequency acquisition under static conditions and high-frequency acquisition under transient conditions. While ensuring complete capture of key transient data, it effectively reduces data redundancy and computational load during static periods, improving the efficiency and relevance of data acquisition and laying the foundation for subsequent accurate prediction.

[0041] For example, the current operating condition type includes at least two of the following: static operating condition, instantaneous operating condition, and steady-state driving condition.

[0042] The data acquisition strategy is dynamically adjusted according to the current operating condition type, including: dynamically adjusting the sampling frequency in response to changes in the current operating condition type; wherein, when in a stationary operating condition, the sampling frequency is reduced; when in a steady-state driving operating condition, the normal sampling frequency is maintained; and when in a transient operating condition, the sampling frequency is increased.

[0043] Further, in one embodiment, the operating parameters include charge and discharge current; identifying the current operating condition type of the battery according to the operating parameters includes: If the absolute value of the charge and discharge current is less than the preset static charge and discharge current threshold and the duration is greater than the preset static time, it is determined as the static operating condition; If the change rate of the charge and discharge current is greater than the preset rapid acceleration change rate, it is determined as the rapid acceleration operating condition under the instantaneous operating condition; If the change rate of the charge and discharge current is less than the preset rapid deceleration change rate, it is determined as the rapid deceleration operating condition under the instantaneous operating condition; If the charge and discharge current is less than 0 and the absolute value of the charge and discharge current is greater than the preset charge and discharge current threshold, it is determined as the charging operating condition under the instantaneous operating condition; it can be understood that the charge and discharge current less than 0 represents charging.

[0044] If it does not meet the static operating condition and the instantaneous operating condition, it is determined as the steady-state driving operating condition.

[0045] In this embodiment, by comprehensively determining the current operating condition type based on the absolute value, change rate, and duration of the charge and discharge current in multiple dimensions, the problems of low accuracy of operating condition identification and incomplete coverage of scenarios caused by traditional single-index judgment are solved. The accurate differentiation of complex operating states such as static, rapid acceleration and deceleration, and charging is realized, ensuring a high degree of matching between the operating condition identification result and the actual physical scenario, providing a reliable basis for the accurate execution of subsequent differential data acquisition strategies, and significantly improving the scenario adaptation ability and operating reliability of the battery management system.

[0046] It can be understood that the above-mentioned static charge and discharge current threshold, preset static time, preset rapid acceleration change rate, preset rapid deceleration change rate, and preset charge and discharge current threshold can all be artificially set according to actual needs, and no specific numerical limitations are made in this application.

[0047] For example, as an example, the charge and discharge current is denoted as I.

[0048] When |I| < I_rest_threshold (for example, 0.5 A) and the duration exceeds t_rest_min (for example, 30 minutes), it is determined as the static operating condition.

[0049] When dI / dt > acc_threshold (for example, 50 A / s), it is determined as the rapid acceleration operating condition.

[0050] When dI / dt < dec_threshold (for example, -50 A / s), it is determined as the rapid deceleration operating condition.

[0051] When I < 0 and |I| > I_charge_threshold (e.g., 5A), it is determined to be in charging condition.

[0052] Other conditions are considered steady-state driving conditions.

[0053] Differentiated token encoding is performed for the different operating conditions identified.

[0054] Specifically, in one embodiment, encoding the state parameters to generate a word sequence includes: when in a static condition, statistically compressing and encoding the state parameters within a statistical time window to generate words representing statistical features, and forming a word sequence from the words in the order of generation; when in a steady-state driving condition or an instantaneous condition, mapping and encoding the state parameters of each sampling point to generate words representing instantaneous features, and forming a word sequence from the words in the order of generation.

[0055] In this embodiment, by adaptively employing statistical compression or mapping encoding to generate lexical sequences based on the operating condition type, the problem of a single encoding strategy being unable to simultaneously achieve data compression efficiency and the integrity of key features is solved. Under static operating conditions, statistical features are used to eliminate temporal redundancy, while under dynamic operating conditions, instantaneous features are used to retain key change details, achieving a balance between data representation efficiency and information density. This not only reduces the computational load on the model but also ensures that the input data accurately reflects the thermal behavior characteristics of the battery under different states, significantly improving the computational efficiency and feature representation capability of subsequent temperature prediction.

[0056] It is understood that the length of the above statistical time window can be set manually according to actual needs, and no specific numerical limit is made in this application.

[0057] For example, in a static operating condition: the monitored temperatures of all monitoring points within the statistical time window are statistically compressed, and the mean, variance, and duration are calculated to generate a macroscopic token, in the following format: [REST|dur=3600s|T_avg=25.2℃|T_std=0.3℃]. This mechanism compresses static data from a statistical time window (lasting for hours or even days) into a single token, significantly reducing the computational load of subsequent prediction models. Understandably, if the state parameters also include other indicators, these indicators are statistically compressed in the same way as the monitored temperatures. For instance, if the state parameters also include operating parameters, such as charging / discharging current, voltage, ambient temperature, and vehicle speed, then each of these four indicators within the statistical time window is statistically compressed, and its mean, variance, and duration are calculated. These values ​​are then combined with the mean, variance, and duration of the monitored temperatures to form a single token.

[0058] For transient conditions: Switch to high-frequency sampling mode (e.g., sampling once every 0.1 seconds), and record the state parameters collected at each sampling point as a token to ensure that temperature spikes and rapid change details are fully captured. Understandably, if the state parameters include other indicators, such as operational parameters, each token will include not only the monitored temperature but also the operational parameters.

[0059] For steady-state driving conditions: using a normal sampling frequency (e.g., 1 second), the state parameters collected at each sampling point are recorded as a token. It is understandable that if the state parameters include other indicators, such as operational parameters, each token will include not only the monitored temperature but also the operational parameters.

[0060] Furthermore, in one embodiment, the method further includes: constructing a directed heat conduction graph G(V,E) with directional nodes and directed edges based on the physical structure of the battery; wherein the physical structure includes the monitoring part, the thermal resistance network inside the battery, and the cooling structure; V represents a node, since a temperature sensor is installed at the monitoring part to measure the temperature of the monitoring part, the node is also the monitoring part or temperature sensor; each node includes a spatial node ID, location description (inside the cell, terminal post, module connection point, heat dissipation channel, etc.), and sensor type; E represents the directed edges between nodes, representing the heat propagation path, and each edge is assigned a heat transfer delay factor, i.e., the theoretical delay time τ.

[0061] Figure 2 This is a visualization of a directed graph representing heat conduction. Nodes in the graph include C1–C6 representing battery cells, T1–T12 representing terminals, B1 and B2 representing module connection points, and D1 representing the heat dissipation channel outlet. Arrows indicate the direction of heat propagation, and the value τ next to each arrow is the theoretical delay time (in seconds). For example, the arrow pointing from cell C1 to terminal T1 is labeled τ=25.3s, indicating that it takes 25.3 seconds for the heat generated by cell C1 to be conducted to terminal T1 and detected by its sensor. The arrow pointing from terminal T1 to module connection point B1 is labeled τ=182.6s, indicating that it takes 182.6 seconds for heat to conduct from the terminal to the module connection point. The heat dissipation channel outlet D1 receives heat from multiple connection points.

[0062] The formula for calculating the theoretical delay time τ is: τ = L 2 / (α·π 2 ), where L is the heat transfer distance (m), i.e., the length of the directed edge, and α is the thermal diffusivity of the material (m). 2 / s), where π is pi. For example, the heat transfer distance L from cell C1 to its positive terminal T1 is 0.005m, and the thermal diffusivity of copper is α≈1.17×10 -4 m2 If the time is τ / s (considering multi-layered materials in actual batteries), then τ≈0.025s. This formula is based on the analytical solution of the one-dimensional heat conduction equation, representing the theoretical time required for heat propagation to reach its peak.

[0063] It should be noted that the above formula is based on a simplified one-dimensional heat conduction model and is used to estimate the characteristic timescale of heat propagation. In actual battery packs, due to the influence of three-dimensional heat transfer, multi-material interfaces, and boundary conditions, the actual heat transfer delay may fluctuate. Therefore, in this application, τ is used as a theoretical reference value, mainly to guide the causal direction and time window range in the sequence construction process, rather than to force precise alignment. The prediction model adaptively learns the optimal time offset through an attention mechanism and training data.

[0064] In this embodiment, a directional heat conduction graph is constructed based on the battery's physical structure, solving the problem of inaccurate prediction of traditional heat propagation paths. This embodiment clearly defines the direction of heat transfer, providing a topological foundation that conforms to physical laws for sequence construction, significantly improving the model's interpretability and prediction accuracy.

[0065] It should be noted that in real battery packs, in addition to solid-state conduction, there are also secondary heat exchange paths such as convection and radiation. Since the heat transfer efficiency of these paths is much lower than that of conduction, and their impact on temperature distribution is small under normal operating conditions, the directed heat conduction graph in this application is mainly constructed based on a solid thermally conductive framework.

[0066] Furthermore, in one embodiment, if the state parameter only includes the monitored temperature, then the lexical sequence only contains the lexical corresponding to the monitored temperature. The lexical sequence is processed using the directed thermal conduction graph of the battery to construct a textual sequence, including: constructing a temperature flow sequence based on the lexical corresponding to the monitored temperature and the directed thermal conduction graph, and using the temperature flow sequence as the textual sequence.

[0067] Furthermore, in one embodiment, the state parameters also include the battery's operating parameters. In this case, the lexical sequence contains not only the lexical corresponding to the monitored temperature but also the lexical corresponding to the operating parameters. The lexical sequence is processed using the directed thermal conduction graph of the battery to construct a textual sequence, including: constructing a temperature flow sequence based on the lexical corresponding to the monitored temperature and the directed thermal conduction graph; and concatenating the temperature flow sequence with the lexical sequence corresponding to the operating parameters to form the textual sequence.

[0068] Furthermore, to enhance the predictive model's ability to learn causal relationships, this application performs time backtracking to reconstruct the input sequence. Specifically, in one embodiment, a temperature flow sequence is constructed based on the terms corresponding to the monitored temperature and the directed heat conduction graph, including: S301: For each directed edge in the heat conduction directed graph, calculate the theoretical delay time.

[0069] S302: For each target time, determine the historical time corresponding to the heat source node on the directed edge based on the theoretical delay time.

[0070] S303: Arrange the tokens corresponding to the monitored temperature of the heat source node at the historical time and the tokens corresponding to the monitored temperature of the heated node on the directed edge at the target time to form a causal sequence block; and splice multiple causal sequence blocks into a temperature flow sequence according to the time order of the target time.

[0071] In this embodiment, by calculating the theoretical delay time of the directed edge, the historical time of the heat source node is determined back based on the target time. The historical terminology of the heat source and the target terminology of the heated element are arranged to form a causal sequence block. This solves the problem that traditional models are difficult to learn due to the neglect of physical heat conduction delay. The model does not need to infer the influence through long-distance dependence. It directly presents the spatiotemporal causal relationship of heat propagation in the input sequence, reduces the difficulty of model learning, and enhances the sensitivity to early signals of heat diffusion. This greatly improves the accuracy and timeliness of battery temperature prediction and provides a reliable basis for thermal runaway early warning.

[0072] As an example, for each monitored temperature, the record format is "[timestamp][spatial node ID][monitored temperature]".

[0073] by Figure 2 Taking a directed heat conduction graph as an example, the temperature flow sequence can be determined by the temporal relationship, for example: [t=0s][C1][47.2℃]→[t=0s][T1][46.8℃]→[t=0s][B1][45.5℃]→[t=0s][D1][44.0℃]…[t=25s ][C1][48.5℃]→[t=25s][T1][47.8℃][τ=25.3s]→[t=25s][B1][46.2℃]→[t=25s][D1][44.5℃] Using the above temporal sequence makes it difficult to enhance the model's ability to learn causal relationships.

[0074] To address this issue, this application employs causal logic to determine the temperature flow sequence, for example: Nodes: C1, T1, B1.

[0075] Heat conduction edge and delay (different edges have different delays): Edge 1 (C1→T1) has a delay of 25 seconds (slow heat transfer), Edge 2 (T1→B1) has a delay of 5 seconds (fast heat transfer).

[0076] Using the target time t as a reference, the historical time of the heat source node is t-τ, thus forming a causal sequence block.

[0077] When the target time t=25s: For edge 1 (C1→T1): the word of the heated node T1 at 25s, the word of the heat source node C1 at 25s-25s=0s, and the causal block 1 is [t=0s][C1][monitoring temperature]→[t=25s][T1][monitoring temperature].

[0078] For edge 2 (T1→B1): the word of the heated node B1 at 25s and the word of the heat source node T1 at 25s-5s=20s are causal block 2 as [t=20s][T1][monitoring temperature]→[t=25s][B1][monitoring temperature].

[0079] The sequence at t=25s is: [t=0s][C1][Monitoring temperature]→[t=25s][T1][Monitoring temperature][τ=25.3s], [t=20s][T1][Monitoring temperature]→[t=25s][B1][Monitoring temperature].

[0080] When the target time t=26s: For edge 1 (C1→T1): the word of the heated node T1 at 26s, the word of the heat source node C1 at 26s-25s=1s, and the causal block 1 is [t=1s][C1][monitoring temperature]→[t=26s][T1][monitoring temperature].

[0081] For edge 2 (T1→B1): the word of the heated node B1 at 26s and the word of the heat source node T1 at 26s-5s=21s are causal block 2 as [t=21s][T1][monitoring temperature]→[t=26s][B1][monitoring temperature].

[0082] The sequence at t=26s is: [t=1s][C1][Monitoring temperature]→[t=26s][T1][Monitoring temperature][τ=25.3s], [t=21s][T1][Monitoring temperature]→[t=26s][B1][Monitoring temperature].

[0083] When the target time t=27s: For edge 1 (C1→T1): the word of the heated node T1 at 27s, the word of the heat source node C1 at 27s-25s=2s, and the causal block 1 is [t=2s][C1][monitoring temperature]→[t=27s][T1][monitoring temperature].

[0084] For edge 2 (T1→B1): the word of the heated node B1 at 27s and the word of the heat source node T1 at 27s-5s=22s are causal block 2 as [t=22s][T1][monitoring temperature]→[t=27s][B1][monitoring temperature].

[0085] The sequence at t=27s is: [t=2s][C1][Monitoring temperature]→[t=27s][T1][Monitoring temperature][τ=25.3s], [t=22s][T1][Monitoring temperature]→[t=27s][B1][Monitoring temperature].

[0086] Concatenating the above time-sequential groups results in the final sequence input to the model: [(t=25s group)[t=0s][C1][Monitoring temperature]→[t=25s][T1][Monitoring temperature],[t=20s][T1][Monitoring temperature]→[t=25s][B1][Monitoring temperature]; (t=26s group) [t=1s][C1][Monitoring temperature]→[t=26s][T1][Monitoring temperature], [t=21s][T1][Monitoring temperature]→[t=26s][B1][Monitoring temperature]; (Group t=27s) [t=2s][C1][Monitored Temperature] → [t=27s][T1][Monitored Temperature], [t=22s][T1][Monitored Temperature] → [t=27s][B1][Monitored Temperature]; ... Further, in one embodiment, arranging the tokens corresponding to the monitored temperature of the heat source node at the historical time and the tokens corresponding to the monitored temperature of the heated node on the directed edge at the target time to form a causal sequence block includes: S401: Calculate the integer number of time steps corresponding to the theoretical delay time.

[0087] For each directed edge E: u→v in the directed graph G(V,E) of heat conduction (where u is the heat source node and v is the heated node), calculate its theoretical delay time τ(u→v).

[0088] Discretize the data sequence over time, with a basic time resolution of t1 (e.g., 1 second). Calculate the integer time steps k = round(τ(u→v) / t1) corresponding to the theoretical delay time.

[0089] S402: Take the nearest time step of the integer time step, and arrange the words corresponding to the monitoring temperature of the heat source node at the historical moment corresponding to the nearest time step in sequence before the words corresponding to the monitoring temperature of the heated node at the target moment to form the causal sequence block.

[0090] In the input sequence, the monitored temperature of the heat source node u at historical time (tk) and the monitored temperature of the heated node v at the target time t are arranged adjacently in the sequence. The specific arrangement format is: [u(tk): monitored temperature] → [v(t): monitored temperature], or the above-mentioned format "[timestamp][spatial node ID][monitored temperature]" can be used.

[0091] To avoid a single k value being too rigid, a backtracking window, or neighboring time step, is actually used. The neighboring time step includes an integer number of time steps and several time steps on at least one side of the integer number of time steps.

[0092] For example, if the target time is 35s, τ(u→v)=30s, t1=1s, then k=30. If the neighboring time steps include an integer time step number and the two time steps to the left of that integer time step number, then the neighboring time steps include 28, 29, and 30, corresponding to historical times of 5s, 6s, and 7s. For example: [u(t-30s)][u(t-29s)][u(t-28s)]→[v(t)].

[0093] For example, if the target time is 35s, τ(u→v)=30s, t1=1s, then k=30. If the neighboring time steps include an integer time step number and the two time steps to the right of that integer time step number, then the neighboring time steps include 30, 31, and 32, corresponding to historical times of 3s, 4s, and 5s. For example: [u(t-32s)][u(t-31s)][u(t-30s)]→[v(t)].

[0094] For example, if the target time is 35s, τ(u→v)=30s, t1=1s, then k=30. If the neighboring time steps include an integer time step number and two time steps to the left and right of that integer time step number, then the neighboring time steps include 28, 29, 30, 31, and 32, and the corresponding historical times are 3s, 4s, 5s, 6s, and 7s.

[0095] When processing sequences, the model does not need to infer "the temperature of a certain node in the past will affect the temperature of another node now" through long-distance dependencies. Instead, it directly sees this causal pairing in the input, which greatly reduces the learning difficulty and improves the sensitivity to early signs of heat diffusion.

[0096] Furthermore, in one embodiment, the prediction model includes an attention mechanism, in which a spatial mask is embedded, comprising: constructing an adjacency matrix, wherein the value corresponding to node pairs with direct directed edges is 1, otherwise it is 0; when calculating attention weights, adding a preset mask value to the scores corresponding to node pairs without direct directed edges, so that the attention weights corresponding to the node pairs without direct directed edges are zero.

[0097] Even with a backtracking window, predictive models, such as the standard Transformer attention mechanism, may still calculate attention weights between all node pairs, including those with no physically direct heat conduction path (e.g., C1 directly with B1). This can cause the model to learn spurious statistical associations, reducing the accuracy of alerts.

[0098] This application embeds a fixed spatial mask matrix M_spatial into the attention mechanism, defined as follows: Construct an adjacency matrix A, where A[i][j]=1 if and only if there exists a directed edge from node i to node j (i→j), or i=j (self-attention is preserved). Otherwise, A[i][j]=0.

[0099] When calculating attention weights, the score matrix Add a mask: S' = S + M_spatial, where: If A[i][j]=1, then M_spatial[i][j]=0; If A[i][j]=0, then M_spatial[i][j]=-∞ (in the implementation, a negative number with a very large absolute value is used, such as -1×10). 9 ).

[0100] Then, a softmax operation is performed on each row of S'. Since the weights of the negative infinity terms approach 0 after softmax, the model is forced to ignore all node pairs without direct heat conduction connections.

[0101] Attention mechanisms can only transmit information along physically permissible heat propagation paths. For example, a model can focus on C1→T1 and T1→B1, but cannot directly focus on C1→B1. Therefore, the effect of heat from C1 to B1 must be transmitted through the intermediate node T1, perfectly aligning with the actual physical process of heat conduction.

[0102] This application uses a backtracking window to solve the "time misalignment" problem: allowing the model to see the direct relationship between "past causes" and "present effects".

[0103] Simultaneously, a spatial mask is used to address the "spatial jump" problem, ensuring that information can only propagate along the actual thermal path. The combination of these two methods allows the predictive model to not only utilize its powerful sequence modeling capabilities but also strictly adhere to the physical laws of heat conduction, thus issuing early warnings within the heat transfer delay window before the heat source causes a significant temperature rise in surrounding sensors.

[0104] In this application, the prediction model uses a pre-trained large time series model.

[0105] The large-scale time series model used in this application is Time-LLMA (based on the LLaMA architecture adapted to time series tasks), and its training process is as follows: Training data: Collect battery temperature data and operating parameters under different battery types (ternary lithium, lithium iron phosphate), different ambient temperatures (-20℃ to 60℃), and different operating conditions (NEDC, WLTC, actual road conditions) to build a training set containing more than 100,000 hours of operating data.

[0106] Data annotation: Annotate the temperature trend within each time window, including normal temperature rise, abnormal temperature rise, and thermal diffusion events.

[0107] Training environment: Pre-training is performed on a cloud GPU cluster (such as NVIDIA A100) that combines supervised learning and self-supervised learning. The loss function is the MSE (mean squared error) between the predicted temperature and the actual measured monitored temperature, plus the cosine similarity loss of the temperature change trend.

[0108] Model deployment: The trained model is compressed and converted into ONNX / TensorRT format and deployed in the vehicle edge computing unit (such as a domain controller), with inference latency controlled within 50ms.

[0109] Further, in one embodiment, the textualized sequence is input into a prediction model to obtain the predicted temperature of each monitoring part of the battery, including: S501: The embedding layer of the prediction model maps the lexical units in the textual sequence into a high-dimensional feature representation of a unified dimension; S502: A prediction input window is constructed based on a preset number of high-dimensional feature representations. The window is then slid across the textual sequence with a sliding step size to make predictions and obtain the predicted temperatures of each monitoring part of the battery in the future time window.

[0110] In this embodiment, by using an embedding layer to map terms of different encoding forms to high-dimensional features of a unified dimension, and by using a sliding window mechanism to perform continuous predictions on the sequence, the problem of difficulty in directly fusing multi-source heterogeneous data and the discrete and inconsistent prediction results is solved. This achieves standardized unification of feature space and continuous temporal extrapolation of data from different operating conditions, ensuring that the model can stably handle mixed encoding sequences, eliminating computational obstacles caused by differences in data dimensionality, significantly improving the continuity and accuracy of temperature prediction, and providing reliable data support for real-time early warning of battery thermal runaway.

[0111] Furthermore, in one embodiment, based on the monitored temperature and the predicted temperature, a temperature warning for the battery is provided, including: S601: Select the monitoring temperature corresponding to the high-dimensional feature representation at the end of the prediction input window, compare it with the predicted temperature at the end of the future time window, and calculate the temperature deviation; S602: Determine the warning level based on the temperature deviation; S603: Execute the corresponding warning according to the warning level.

[0112] In this embodiment, the deviation between the monitored temperature at the end of the predicted input window and the predicted temperature at the end of a future time window is calculated, and the warning level is determined based on the deviation to execute the warning. This solves the problems of traditional fixed threshold warnings being lagging and unable to distinguish the degree of risk. It achieves hierarchical forward-looking warnings based on prediction deviation, enabling the identification of abnormal trends before the temperature exceeds the limit, distinguishing between minor deviations and serious risks, significantly improving the timeliness and accuracy of warnings, avoiding false alarms and missed alarms, providing differentiated control basis for battery thermal management, and ensuring the safe operation of the battery.

[0113] Furthermore, the early warning system is categorized using temperature deviation ΔT, as detailed in Table 1 below.

[0114] Furthermore, the rate of temperature change can also be calculated. Graded early warning is based on the rate of temperature change.

[0115] in, The duration of the future time window can be determined based on actual needs. The predicted temperature at the end of the future time window is the last predicted temperature output after prediction, which is the predicted value. The predicted temperature at the end of the input window, i.e., the last monitored temperature in the input data, is the measured value.

[0116] Furthermore, temperature acceleration can also be calculated. : .

[0117] Furthermore, spatial thermal equilibrium can be assessed: based on the directed heat conduction graph, the temperature difference between different nodes is calculated and compared with the theoretical thermal equilibrium threshold, as shown in Table 1 below.

[0118] Furthermore, heat transfer delay anomalies can be assessed: based on the statistical distribution of heat transfer delays between node pairs learned by the prediction model under normal operating conditions (e.g., mean μ_delay and standard deviation σ_delay), the deviation between the currently observed node pair delay time and the delay time of this statistical distribution is calculated. If this deviation exceeds a preset multiple (e.g., μ_delay ± 3σ_delay), it is determined to be an abnormal heat conduction path. The theoretical delay τ can be used as an initialization or reference baseline, but not as the sole criterion. See Table 1 below for details.

[0119] When making a judgment, an early warning can be issued when the predicted temperature of any monitoring point triggers the corresponding condition.

[0120] Table 1

[0121] The methods for determining the first, second, and third deviation thresholds and the first and second rate thresholds can be as follows: based on the 95th, 99th, and 99.9th percentile statistics of historical normal operating data, and can be dynamically calibrated as the battery ages. Other judgment thresholds, such as acceleration threshold, delay time deviation, and theoretical thermal equilibrium threshold, can be set according to actual needs.

[0122] After implementing the early warning strategy, multiple systems work together.

[0123] Linked with the vehicle's air conditioning system: Automatically adjusts the air conditioning's cooling capacity and ventilation mode, directing cool air to the battery's heat dissipation channel.

[0124] Linked with the vehicle control system: Limits the maximum vehicle speed and prohibits rapid acceleration / deceleration.

[0125] Link with remote monitoring center: Upload early warning events and battery status data, and receive remote commands.

[0126] Integration with mobile devices: Push warning information and response suggestions to car owners via the application (such as "Please park in a safe place").

[0127] In summary, this application achieves at least the following advantages: Dynamically adapting to operating conditions while balancing efficiency and accuracy: Through dynamic slicing driven by operating conditions, long-term static data is compressed into macro-level tokens, significantly reducing the computing power overhead for edge inference; at the same time, high-frequency encryption of transient operating conditions ensures that key thermal evolution details are not lost.

[0128] Physically embedded spatial topology enables early warning of heat diffusion paths: For the first time, a directed graph of battery heat conduction is embedded into the feature space of a large-scale temporal model, enabling the model to learn the causal relationship of heat propagation in space. Experiments show that this application can issue an early warning within the heat transfer delay window (approximately 15-30 seconds in advance) after the occurrence of a local heat source and before the temperature of surrounding sensors rises significantly, far exceeding traditional methods based on difference thresholds.

[0129] Multi-dimensional feature fusion hierarchical early warning: By integrating temperature deviation, rate of change, acceleration and spatial thermal equilibrium indicators, a refined four-level early warning is achieved, forming a complete closed loop from risk identification to automated intervention.

[0130] In summary, this application solves the technical challenges of "early identification of heat diffusion paths" and "the contradiction between long-cycle redundancy and transient loss" that cannot be addressed by existing technologies through a hybrid enhanced intelligence solution of "physical embedding combined with a large temporal model".

[0131] Secondly, embodiments of this application also provide a battery temperature management device.

[0132] In one embodiment, reference is made to Figure 3 , Figure 3 This is a functional module diagram of an embodiment of the battery temperature management device of this application. Figure 3 As shown, the battery temperature management device includes: The parameter acquisition module is used to acquire the state parameters of the battery and encode the state parameters to generate a word sequence. The state parameters include the monitoring temperature of each monitoring part of the battery. A text generation module is used to process the word sequence using the directed graph of thermal conduction of the battery to construct a textual sequence; A temperature prediction module is used to input the textual sequence into the prediction model to obtain the predicted temperature of each monitoring part of the battery. An early warning module is used to provide temperature warnings for the battery based on the monitored temperature and the predicted temperature.

[0133] Furthermore, in one embodiment, the parameter acquisition module acquires the battery's status parameters, including: collecting the battery's operating parameters; identifying the battery's current operating condition type based on the operating parameters; dynamically adjusting the data acquisition strategy based on the current operating condition type; and acquiring the status parameters based on the adjusted data acquisition strategy.

[0134] Furthermore, in one embodiment, the parameter acquisition module dynamically adjusts the data acquisition strategy according to the current operating condition type, including: dynamically adjusting the sampling frequency in response to changes in the current operating condition type; wherein, when in a stationary operating condition, the sampling frequency is reduced; when in a steady-state driving operating condition, the normal sampling frequency is maintained; and when in a transient operating condition, the sampling frequency is increased.

[0135] Further, in one embodiment, the parameter acquisition module identifies the current operating condition type of the battery based on the operating parameters, including: if the absolute value of the charging and discharging current is less than a preset static charging and discharging current threshold and the duration is greater than a preset static time, it is determined to be a static operating condition; if the rate of change of the charging and discharging current is greater than a preset rapid acceleration rate of change, it is determined to be a rapid acceleration operating condition under an instantaneous operating condition; if the rate of change of the charging and discharging current is less than a preset rapid deceleration rate of change, it is determined to be a rapid deceleration operating condition under an instantaneous operating condition; if the charging and discharging current is less than 0 and the absolute value of the charging and discharging current is greater than a preset charging and discharging current threshold, it is determined to be a charging operating condition under an instantaneous operating condition; if none of these conditions meet the requirements of the static and instantaneous operating conditions, it is determined to be a steady-state driving operating condition.

[0136] Furthermore, in one embodiment, the parameter acquisition module encodes the state parameters to generate a word sequence, including: when in a static condition, statistically compressing and encoding the state parameters within a statistical time window to generate words representing statistical features, and the words forming a word sequence in the order of generation; when in a steady-state driving condition or an instantaneous condition, mapping and encoding the state parameters of each sampling point to generate words representing instantaneous features, and the words forming a word sequence in the order of generation.

[0137] Furthermore, in one embodiment, the battery temperature management device further includes a directed graph construction module, which is used to construct a directed heat conduction graph with directional nodes and directed edges based on the physical structure of the battery; wherein the physical structure includes the monitoring part, the thermal resistance network inside the battery, and the cooling structure.

[0138] Furthermore, in one embodiment, the text generation module processes the lexical sequence using the directed thermal conduction graph of the battery to construct a textual sequence, including: constructing a temperature flow sequence based on the lexical corresponding to the monitored temperature and the directed thermal conduction graph, and using the temperature flow sequence as the textual sequence.

[0139] Furthermore, in one embodiment, the text generation module processes the lexical sequence using the directed thermal conduction graph of the battery to construct a textual sequence, including: constructing a temperature flow sequence based on the lexical corresponding to the monitored temperature and the directed thermal conduction graph; and concatenating the temperature flow sequence with the lexical sequence corresponding to the operating parameters to form the textual sequence.

[0140] Further, in one embodiment, the text generation module constructs a temperature flow sequence based on the lexical units corresponding to the monitored temperature and the directed graph of heat conduction, including: calculating the theoretical delay time for each directed edge in the directed graph of heat conduction; determining the historical time corresponding to the heat source node on the directed edge based on the theoretical delay time for each target time; arranging the lexical units corresponding to the monitored temperature of the heat source node at the historical time and the lexical units corresponding to the monitored temperature of the heated node on the directed edge at the target time to form a causal sequence block; and concatenating multiple causal sequence blocks in chronological order of the target time to form a temperature flow sequence.

[0141] Further, in one embodiment, the text generation module arranges the tokens corresponding to the monitored temperature of the heat source node at the historical time and the tokens corresponding to the monitored temperature of the heated node at the target time on the directed edge to form a causal sequence block, including: calculating the integer time step corresponding to the theoretical delay time; taking the nearest time step of the integer time step, and sequentially arranging the tokens corresponding to the monitored temperature of the heat source node at the historical time corresponding to the nearest time step before the tokens corresponding to the monitored temperature of the heated node at the target time to form the causal sequence block.

[0142] Furthermore, in one embodiment, the prediction model includes an attention mechanism, in which a spatial mask is embedded, comprising: constructing an adjacency matrix, wherein the value corresponding to node pairs with direct directed edges is 1, otherwise it is 0; when calculating attention weights, adding a preset mask value to the scores corresponding to node pairs without direct directed edges, so that the attention weights corresponding to the node pairs without direct directed edges are zero.

[0143] Furthermore, in one embodiment, the temperature prediction module inputs the textual sequence into a prediction model to obtain the predicted temperature of each monitoring part of the battery, including: mapping the lexical units in the textual sequence into high-dimensional feature representations of a uniform dimension through the embedding layer of the prediction model; constructing a prediction input window based on a preset number of high-dimensional feature representations, and sliding it on the textual sequence with a sliding step size to make predictions, thereby obtaining the predicted temperature of each monitoring part of the battery in a future time window.

[0144] Furthermore, in one embodiment, the early warning module provides a temperature warning for the battery based on the monitored temperature and the predicted temperature, including: selecting the monitored temperature corresponding to the high-dimensional feature representation at the end of the predicted input window, comparing it with the predicted temperature at the end of the future time window, and calculating the temperature deviation; determining the early warning level based on the temperature deviation; and executing the corresponding early warning based on the early warning level.

[0145] The functions of each module in the battery temperature management device correspond to the steps in the battery temperature management method embodiment, and their functions and implementation processes will not be described in detail here.

[0146] Thirdly, embodiments of this application provide a battery temperature management device, which can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.

[0147] Reference Figure 4 , Figure 4 This is a schematic diagram of the hardware structure of the battery temperature management device involved in the embodiments of this application. In the embodiments of this application, the battery temperature management device may include a processor, a memory, a communication interface, and a communication bus.

[0148] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0149] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting devices within the battery temperature management device, as well as interfaces used for interconnecting the battery temperature management device with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0150] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0151] The processor can be a general-purpose processor, which can call the battery temperature management program stored in the memory and execute the battery temperature management method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the battery temperature management program is called can be referred to in the various embodiments of the battery temperature management method of this application, and will not be repeated here.

[0152] Those skilled in the art will understand that Figure 4 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0153] Fourthly, embodiments of this application also provide a computer-readable storage medium.

[0154] The present application has a computer-readable storage medium storing a battery temperature management program, wherein when the battery temperature management program is executed by a processor, it implements the steps of the battery temperature management method described above.

[0155] The method implemented when the battery temperature management program is executed can be referred to in various embodiments of the battery temperature management method of this application, and will not be repeated here.

[0156] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0157] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0158] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0159] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0160] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0161] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0162] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A battery temperature management method, characterized in that, include: The battery status parameters are acquired and encoded to generate a word sequence. The status parameters include the monitoring temperature of each monitoring part of the battery. For each directed edge in the directed graph of the battery's thermal conduction, the theoretical delay time is calculated; for each target time, the historical time corresponding to the heat source node on the directed edge is determined based on the theoretical delay time; the integer time step corresponding to the theoretical delay time is calculated; the nearest time step of the integer time step is taken, and the tokens corresponding to the monitored temperature of the heat source node at the historical time corresponding to the nearest time step are sequentially arranged before the tokens corresponding to the monitored temperature of the heated node at the target time on the directed edge, forming a causal sequence block; multiple causal sequence blocks are concatenated into a temperature flow sequence according to the time order of the target time; the temperature flow sequence is used as a textual sequence. The textual sequence is input into the prediction model to obtain the predicted temperature of each monitoring part of the battery; Based on the monitored and predicted temperatures, a temperature warning is issued for the battery.

2. The battery temperature management method as described in claim 1, characterized in that, Obtain the battery's status parameters, including: Collect the operating parameters of the battery; The current operating condition type of the battery is identified based on the operating parameters; The data acquisition strategy is dynamically adjusted based on the current operating condition type. The state parameters are collected based on the adjusted data acquisition strategy.

3. The battery temperature management method as described in claim 2, characterized in that: The current operating condition type includes at least two of the following: static operating condition, instantaneous operating condition, and steady-state driving condition; Dynamically adjust the data acquisition strategy according to the current operating condition type, including: The sampling frequency is dynamically adjusted in response to changes in the current operating condition type. Specifically, when the vehicle is stationary, the sampling frequency is reduced; when the vehicle is in steady-state driving condition, the normal sampling frequency is maintained; and when the vehicle is in transient condition, the sampling frequency is increased.

4. The battery temperature management method as described in claim 2, characterized in that, The operating parameters include charging and discharging current; identifying the current operating condition type of the battery based on the operating parameters includes: If the absolute value of the charging and discharging current is less than the preset static charging and discharging current threshold and the duration is greater than the preset static time, it is determined to be a static working condition. If the rate of change of the charging and discharging current is greater than the preset rapid acceleration rate of change, it is determined to be a rapid acceleration condition under instantaneous operating conditions. If the rate of change of the charging and discharging current is less than the preset rate of change of rapid deceleration, it is determined to be a rapid deceleration condition under instantaneous operating conditions. If the charging and discharging current is less than 0 and the absolute value of the charging and discharging current is greater than the preset charging and discharging current threshold, it is determined to be a charging condition under instantaneous operating conditions. If neither of the above conditions meets the requirements for static or instantaneous operation, the condition is determined to be a steady-state driving condition.

5. The battery temperature management method as described in claim 2, characterized in that: The current operating condition type includes at least two of the following: static operating condition, instantaneous operating condition, and steady-state driving condition; Encoding the state parameters to generate a word sequence includes: When in a static operating condition, the state parameters within the statistical time window are statistically compressed and encoded to generate lexical units that represent statistical features, and the lexical units are arranged in the order of generation to form a lexical unit sequence. When in steady-state or instantaneous driving conditions, the state parameters of each sampling point are mapped and encoded to generate lexical units that represent instantaneous features. The lexical units are then arranged in the order of generation to form a lexical unit sequence.

6. The battery temperature management method as described in claim 1, characterized in that, The method further includes: Based on the physical structure of the battery, a directed heat conduction graph with directional nodes and directed edges is constructed; The physical structure includes the monitoring part, the thermal resistance network inside the battery, and the cooling structure.

7. The battery temperature management method as described in claim 1, characterized in that: The status parameters also include the battery's operating parameters; The temperature flow sequence is concatenated with the word sequence corresponding to the operating parameters to form the textual sequence.

8. The battery temperature management method as described in claim 1, characterized in that: The prediction model includes an attention mechanism, in which a spatial mask is embedded, including: Construct an adjacency matrix where nodes with direct directed edges are assigned a value of 1, and nodes without direct directed edges are assigned a value of 0. When calculating attention weights, a preset mask value is added to the scores of nodes that do not have direct directed edges, so that the attention weights of the nodes that do not have direct directed edges are zero.

9. The battery temperature management method as described in claim 1, characterized in that: The textual sequence is input into the prediction model to obtain the predicted temperature of each monitoring part of the battery, including: The embedding layer of the prediction model maps the lexical units in the textual sequence into a high-dimensional feature representation of a unified dimension. A prediction input window is constructed based on a preset number of high-dimensional feature representations. The window slides on the textual sequence with a sliding step size to make predictions, thereby obtaining the predicted temperature of each monitoring part of the battery in the future time window.

10. The battery temperature management method as described in claim 9, characterized in that: Based on the monitored and predicted temperatures, a temperature warning is issued for the battery, including: The high-dimensional feature representation corresponding to the end of the prediction input window is selected and compared with the predicted temperature at the end of the future time window to calculate the temperature deviation. The warning level is determined based on the temperature deviation. The corresponding warning will be issued based on the warning level.

11. A battery temperature management device, characterized in that, include: The parameter acquisition module is used to acquire the state parameters of the battery and encode the state parameters to generate a word sequence. The state parameters include the monitoring temperature of each monitoring part of the battery. A text generation module is used to calculate the theoretical delay time for each directed edge in the directed graph of the battery's thermal conduction; for each target time, determine the historical time corresponding to the heat source node on the directed edge based on the theoretical delay time; calculate the integer time step corresponding to the theoretical delay time; take the nearest time step of the integer time step, and arrange the words corresponding to the monitored temperature of the heat source node at the historical time corresponding to the nearest time step in sequence before the words corresponding to the monitored temperature of the heated node at the target time on the directed edge, forming a causal sequence block; splice multiple causal sequence blocks into a temperature flow sequence according to the time order of the target time; and use the temperature flow sequence as a textualized sequence. A temperature prediction module is used to input the textual sequence into the prediction model to obtain the predicted temperature of each monitoring part of the battery. An early warning module is used to provide temperature warnings for the battery based on the monitored temperature and the predicted temperature.

12. A battery temperature management device, characterized in that, The battery temperature management device includes a processor, a memory, and a battery temperature management program stored in the memory and executable by the processor, wherein when the battery temperature management program is executed by the processor, it implements the steps of the battery temperature management method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a battery temperature management program, wherein when the battery temperature management program is executed by a processor, it implements the steps of the battery temperature management method as described in any one of claims 1 to 10.

Images