Battery thermal management method, device, apparatus and readable storage medium

Abstract

The application discloses a battery thermal management method, device and equipment and a readable storage medium, relates to the technical field of battery management, and comprises the following steps: acquiring monitoring data of a key part of a battery; constructing a time-space sensing data matrix according to the monitoring data, and generating a battery thermal runaway feature sequence; determining hotspot temperature prediction data and thermal runaway residual time prediction data according to the battery thermal runaway feature sequence and a feature decoding parameter set constructed by pre-fitting; calibrating the hotspot temperature prediction data and the thermal runaway residual time prediction data according to a historical hotspot temperature calibration set and a historical thermal runaway residual time calibration set to obtain an upper confidence boundary set, a lower confidence boundary set and an interval width set; generating a battery thermal runaway prediction result according to the upper confidence boundary set, the lower confidence boundary set and the interval width set, and then outputting a battery thermal management control instruction. The application solves the problems of low thermal runaway alarm credibility and thermal runaway early warning lag of the existing battery thermal management method.

Description

Technical Field

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

[0002] With the rapid development of new energy vehicles and energy storage power stations, the safety of lithium-ion batteries has received increasing attention. Battery thermal runaway is one of the main causes of battery safety accidents. Therefore, building an efficient and accurate battery thermal management system is crucial for preventing thermal runaway accidents.

[0003] In existing technologies, battery thermal management mainly relies on the battery management system (BMS) to collect the surface temperature of the battery cells and to alarm and control the system based on preset temperature thresholds. That is, when the monitored temperature value exceeds the preset temperature threshold, the cooling system of the entire battery pack is activated or the main power supply is cut off.

[0004] However, the above-mentioned solution is based on a preset temperature threshold for alarm, which is prone to false alarms or missed alarms due to sensor noise, and lacks effective early warning, affecting the system's energy efficiency and safety. Summary of the Invention

[0005] This invention provides a battery thermal management method, apparatus, device, and readable storage medium to solve the technical problems of low reliability and delayed early warning in the prior art where battery thermal runaway alarms are based on preset temperature thresholds.

[0006] Firstly, a battery thermal management method is provided, comprising the following steps: Acquire monitoring data for key components of the battery, including cell tab monitoring temperature, cell surface monitoring temperature, battery surface monitoring pressure, and busbar monitoring strain; A spatiotemporal sensing data matrix is ​​constructed based on the monitoring data, and a battery thermal runaway characteristic sequence is generated based on the spatiotemporal sensing data matrix. Based on the battery thermal runaway feature sequence and the pre-fitted feature decoding parameter set, determine the hot spot temperature prediction data and the thermal runaway remaining time prediction data; The hotspot temperature prediction data and the thermal runaway remaining time prediction data are calibrated based on the historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set to obtain the upper confidence boundary set, the lower confidence boundary set and the interval width set. The battery thermal runaway prediction results are generated based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set. The battery thermal management control command is then output based on the battery thermal runaway prediction results.

[0007] In some embodiments, the step of constructing a spatiotemporal sensing data matrix based on the monitoring data and generating a battery thermal runaway feature sequence based on the spatiotemporal sensing data matrix includes: The monitoring data is numerically normalized, and the normalized monitoring data is aligned according to the time stamp order to construct a spatiotemporal sensing data matrix. The spatiotemporal sensing data matrix is ​​input into a pre-trained deep learning model to obtain spatial coupling weights and temporal correlation weights. The spatiotemporal sensing data matrix is ​​then weighted and fused based on the spatial coupling weights and temporal correlation weights. Based on the weighted fusion spatiotemporal sensing data matrix, the busbar strain response identifier, the pressure change and heat accumulation abnormal coupling identifier, the cell tab area temperature rise slope abnormal identifier, and the cell surface temperature diffusion abnormal identifier are determined, and they are aligned in chronological order according to the timestamps to generate the battery thermal runaway feature sequence.

[0008] In some embodiments, determining the hotspot temperature prediction data and the remaining thermal runaway time prediction data based on the battery thermal runaway feature sequence and a pre-fitted feature decoding parameter set includes: Extract the decoding weight matrix and decoding bias vector from the pre-fitted feature decoding parameter set, and merge the decoding weight matrix and decoding bias vector with the battery thermal runaway feature sequence to obtain the hidden feature vector; Temperature prediction regression vector and time prediction regression vector are extracted from the pre-fitted feature decoding parameter set. The inner product value between the hidden feature vector and the temperature prediction regression vector is calculated to obtain hotspot temperature prediction data. The inner product value between the hidden feature vector and the time prediction regression vector is also calculated to obtain thermal runaway remaining time prediction data.

[0009] In some embodiments, the calibration of the hotspot temperature prediction data and the thermal runaway remaining time prediction data based on the historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set to obtain the upper confidence boundary set, the lower confidence boundary set, and the interval width set includes: The historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set are arranged in descending order. Based on the preset distribution quantile, the corresponding historical error quantile values ​​are extracted from the historical temperature calibration set and the historical remaining time calibration set after descending order as compensation range parameters to establish the hotspot temperature prediction calibration margin and the thermal runaway remaining time prediction calibration margin. Calculate the summative and subtractive extreme values ​​of the hotspot temperature prediction data and the thermal runaway remaining time prediction data with the corresponding hotspot temperature prediction calibration margin and thermal runaway remaining time prediction calibration margin, respectively. Merge each summative extreme value to obtain the upper confidence boundary set and each subtractive extreme value to obtain the lower confidence boundary set. Calculate the extreme value deviation between the corresponding values ​​of the boundary sets to obtain the interval width set.

[0010] In some embodiments, generating battery thermal runaway prediction results based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set includes: The extreme values ​​of the summation distribution of each item in the upper confidence boundary set are compared with the preset temperature risk judgment threshold, the extreme values ​​of the subtraction distribution of each item in the lower confidence boundary set are compared with the preset time emergency judgment threshold, and the interval width set is compared with the preset anomaly amplification width threshold to obtain the battery thermal runaway prediction result.

[0011] In some embodiments, the step of outputting battery thermal management control commands based on battery thermal runaway prediction results includes: The system outputs a battery thermal management control command to perform directional physical cooling on the battery risk sites indicated by the battery thermal runaway prediction results, and / or outputs a battery thermal management control command to regulate the current in the current transmission path corresponding to the battery risk sites indicated by the battery thermal runaway prediction results.

[0012] In some embodiments, acquiring monitoring data of key battery components includes: By attaching a flexible printed circuit board to key parts of the battery, the induced electrical signals from temperature sensors, pressure sensors, and strain sensors integrated on the flexible printed circuit board are acquired, and monitoring data of key parts of the battery are obtained based on the induced electrical signals.

[0013] Secondly, a battery thermal management device is provided, comprising: The acquisition unit is used to acquire monitoring data of key parts of the battery, including the monitoring temperature of the cell tabs, the monitoring temperature of the cell surface, the monitoring pressure of the battery surface, and the monitoring strain of the busbar. The construction unit is used to construct a spatiotemporal sensing data matrix based on the monitoring data, and generate a battery thermal runaway feature sequence based on the spatiotemporal sensing data matrix. The determining unit is used to determine the hot spot temperature prediction data and the thermal runaway remaining time prediction data based on the battery thermal runaway feature sequence and the pre-fitted feature decoding parameter set. The calibration unit is used to calibrate the hot spot temperature prediction data and the thermal runaway remaining time prediction value according to the historical hot spot temperature calibration set and the historical thermal runaway remaining time calibration set, to obtain the upper confidence boundary set, the lower confidence boundary set and the interval width set. The output unit is used to generate battery thermal runaway prediction results based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set, and to output battery thermal management control commands based on the battery thermal runaway prediction results.

[0014] Thirdly, a computer device is provided, comprising: a memory and a processor, wherein the memory stores at least one instruction, the at least one instruction being loaded and executed by the processor to implement the aforementioned battery thermal management method.

[0015] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the aforementioned battery thermal management method.

[0016] The beneficial effects of the technical solution provided by this invention include: This invention provides a battery thermal management method, apparatus, device, and readable storage medium. It constructs a spatiotemporal sensing data matrix by acquiring multi-physics field monitoring data, including cell tab temperature, surface pressure, and busbar strain. It then predicts hotspot temperatures and remaining thermal runaway time by combining feature decoding parameters. Furthermore, it calculates the upper / lower confidence boundaries and interval width of the predicted values ​​using historical calibration errors. Finally, it generates thermal management control commands based on the set of confidence boundaries. This method captures early signs through multi-physics field coupling features, providing early warning capabilities for thermal runaway. The confidence boundaries enhance alarm reliability, solving the technical problems of low reliability and delayed thermal runaway warnings caused by alarms based on preset temperature thresholds in existing technologies. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of a battery thermal management method provided in an embodiment of the present invention; Figure 2 Provided for embodiments of the present invention Figure 1 A flowchart illustrating step S20 in the implementation process; Figure 3 Provided for embodiments of the present invention Figure 1 A flowchart illustrating step S30 in the implementation process; Figure 4 Provided for embodiments of the present invention Figure 1 A flowchart illustrating step S40 in the implementation process; Figure 5 This is a schematic diagram of the structure of a battery thermal management device provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention provides a battery thermal management method that can solve the technical problems of low alarm reliability and lack of early warning capability in the prior art, which relies on preset temperature thresholds for alarm and control of battery thermal runaway.

[0021] like Figure 1 As shown, this embodiment of the invention provides a battery thermal management method, including the following steps: Step S100: Real-time acquisition of monitoring data for key parts of the battery. The monitoring data includes the cell tab monitoring temperature sequence, the cell surface monitoring temperature sequence, the battery surface monitoring pressure sequence, and the busbar monitoring strain sequence.

[0022] Specifically, by attaching a flexible printed circuit board to key parts of the battery, the induced electrical signals (usually voltage signals) from temperature sensors, pressure sensors, and strain sensors integrated on the flexible printed circuit board are acquired, and monitoring data of key parts of the battery are obtained based on the induced electrical signals.

[0023] Distributed temperature sensors, distributed pressure sensors, and distributed strain sensors are integrated onto a flexible printed circuit board (PCB). This PCB acts as the nerve endings, attached to key points on critical battery components such as the cell tabs, cell surface, and busbars. This enables continuous, uninterrupted real-time, blind-zone-free in-situ status monitoring, resulting in a multi-node in-situ monitoring coordinate array. The specific operation process is as follows: First, measure the actual dimensions of the battery cell tab length, tab width, surface length, surface width, and the busbar bending and connecting sections. Cut the flexible printed circuit board (FPC) according to the measured dimensions, ensuring the outer edge of the board aligns with the battery cell tabs, surface, and busbar attachment area. Next, solder distributed temperature sensors, distributed pressure sensors, and distributed strain sensors onto the board at predetermined locations, writing a unique address code for each sensor node. Then, write the node address code, horizontal coordinate, vertical coordinate, and component position into the data acquisition configuration table. Subsequently, cleaning, positioning, and pressing were performed on the cell tabs, cell surface, and busbar attachment area. Temperature sensing nodes were arranged along the current flow direction in the cell tab area, along the length and width directions on the cell surface, pressure sensing nodes were arranged in the stress areas of the battery surface, and strain sensing nodes were arranged in the busbar bending area, around the connection holes, and in the current concentration area. The pressure applied by the pressure head was recorded during pressing. When the pressing pressure was between 0.15 MPa and 0.30 MPa, the adhesive layer spread continuously and the initial output of the sensitive layer was stable, which was recorded as effective adhesion. Below 0.15 MPa... The adhesive layer was not compacted and the node output was prone to drift, which was written as unstable contact. When the pressure was higher than 0.30 MPa, the sensitive layer was deformed under pressure and the zero-point offset increased, which was written as local compression. Then, the acquisition hardware sent trigger signals to each node in sequence, and read the node address, sampling time and original electrical signal value point by point. The sampling period was set to 10 ms to 100 ms. When it was lower than 10 ms, the increment between adjacent sampling points was close to noise fluctuation and the repeatability of the recorded values ​​was too high, which was written as time redundancy. When it was higher than 100 ms, the local temperature rise and strain change process would span too large an interval, which was written as insufficient time resolution. Finally, all node records were written to the acquisition file in chronological order, and the node address, installation coordinates, component type, original electrical signal value and timestamp were imported into the tool software (such as MATLAB). All records in the continuous time period were stacked in chronological order. The recording result is a multi-node in-situ monitoring coordinate array corresponding to the key points of the cell tab, cell surface, battery surface and busbar.

[0024] Next, extract the temperature values ​​at the cell tabs and output the real-time monitoring temperature sequence of the cell tabs; extract the temperature values ​​at the cell surface and output the real-time monitoring temperature sequence of the cell surface; extract the pressure values ​​at the battery surface and output the real-time monitoring pressure sequence of the battery surface; extract the strain values ​​at the busbar and output the real-time monitoring strain sequence of the busbar. The specific process is as follows: Temperature nodes at the cell tab locations are selected from the multi-node in-situ monitoring coordinate array based on component labels. The original resistance or voltage value of each node is read, and then the reference temperature value, reference electrical signal value, and temperature sensitivity from the node's calibration file are retrieved. The calibration file is formed by pairing temperature and electrical signal values ​​recorded point-by-point at multiple known temperature points. The current temperature value = calibration reference temperature value + (current electrical signal value - calibration reference electrical signal value) / temperature sensitivity. Subsequently, the temperature values ​​at the same cell tab location at different sampling times are written into the same column in chronological order, and the temperature values ​​at different cell tab locations at the same time are written into the same row in spatial coordinate order, forming a real-time monitoring temperature sequence for the cell tabs. Then, the same steps are followed to select cell surface temperature nodes and perform temperature conversion. The resulting cell surface temperature values ​​are arranged in chronological and spatial coordinate order, forming a real-time monitoring temperature sequence for the cell surface. Next, pressure nodes on the battery surface are identified, and the bridge output voltage, zero-point bridge output voltage, and pressure sensitivity are read. The pressure sensitivity is calculated by recording the bridge output voltage at multiple known pressure load points for that node. The current pressure value is calculated as (current bridge output voltage - zero-point bridge output voltage) / pressure sensitivity. When the pressure value is between 0 kPa and 5 kPa, it indicates that the outer surface of the battery is only subjected to slight contact load, which is written as the low-pressure range. When it is between 5 kPa and 30 kPa, it indicates that the outer surface of the battery is subjected to normal loads caused by assembly clamping and running vibration, which is written as the medium-pressure range. When it is above 30 kPa, it indicates localized stress concentration, which is written as the high-pressure range. Subsequently, the pressure values ​​at the same battery surface stress location at different sampling times are written into the same column in chronological order. The pressure values ​​at different battery surface pressure nodes at the same time are written into the same row in installation coordinate order, and the corresponding low-pressure range, medium-pressure range, or high-pressure range identifiers are retained simultaneously to form a real-time monitoring pressure sequence of the battery surface. Finally, the strain nodes of the busbars were screened out, and the output voltage, zero-point voltage, excitation voltage, and strain gauge sensitivity coefficient of the bridge were read. The strain gauge sensitivity coefficient was determined by applying known displacements to the nodes step by step on a standard bending specimen and recording the bridge output. The current strain value = current bridge output voltage deviation / (strain gauge sensitivity coefficient × excitation voltage). When the absolute strain value is between 0 microstrain and 300 microstrain, it indicates that the busbar is in the range of small elastic deformation, which is written as the low deformation range. When it is between 300 microstrain and 1200 microstrain, it indicates that the connection part has shown obvious bending or tensile and compressive changes, which is written as the medium deformation range. When it is higher than 1200 microstrain, it indicates that the local deformation amplitude has increased significantly, which is written as the high deformation range. Subsequently, the strain values ​​of the same busbar bending area, around the connection hole, or the flow concentration area at different sampling times were written into the same column in chronological order. The strain values ​​of different busbar strain nodes at the same time were written into the same row in the order of installation coordinates, and the corresponding low deformation range, medium deformation range, or high deformation range identifiers were retained simultaneously to form a real-time monitoring strain sequence of the busbars.After the above processing, we obtain the real-time monitoring temperature sequence of the cell tab, the real-time monitoring temperature sequence of the cell surface, the real-time monitoring pressure sequence of the battery surface, and the real-time monitoring strain sequence of the busbar.

[0025] Step S200: Construct a spatiotemporal sensing data matrix based on the monitoring data, and generate a battery thermal runaway characteristic sequence based on the spatiotemporal sensing data matrix.

[0026] Specifically, such as Figure 2 As shown, a spatiotemporal sensing data matrix is ​​constructed based on monitoring data, and a battery thermal runaway characteristic sequence is generated based on the spatiotemporal sensing data matrix, including: Step S201: Perform numerical normalization processing on the multiphysics monitoring data, align the numerically normalized multiphysics monitoring data according to the time stamp order, and construct a spatiotemporal sensing data matrix.

[0027] Within the current processing window, read the maximum, minimum, and current values ​​of the real-time monitoring temperature sequence of the cell tab, the real-time monitoring temperature sequence of the cell surface, the real-time monitoring pressure sequence of the battery surface, and the real-time monitoring strain sequence of the busbar. Then, perform extreme value normalization mapping point by point on the four types of sequences: Normalized temperature value of cell tab = (current cell tab temperature value - minimum cell tab temperature value) / (maximum cell tab temperature value - minimum cell tab temperature value); Normalized temperature value of cell surface = (current cell surface temperature value - minimum cell surface temperature value) / (maximum cell surface temperature value - minimum cell surface temperature value); Normalized pressure value of battery surface = (current battery surface pressure value - minimum battery surface pressure value) / (maximum battery surface pressure value - minimum battery surface pressure value); Normalized strain value of busbar = (current busbar strain value - minimum busbar strain value) / (maximum busbar strain value - minimum busbar strain value). After the above processing, the normalized temperature sequence of the cell tab, the normalized temperature sequence of the cell surface, the normalized pressure sequence of the battery surface, and the normalized strain sequence of the busbar are obtained. When the maximum and minimum values ​​of any sequence are equal, it indicates that the sequence has no expansion range within that window and cannot form a proportional distribution; the normalization result of this sequence within the current window is written as 0. 0 is used as the lower limit of normalization because when the current value equals the minimum value of the window, it is not higher than the lowest level of this window. 1 is used as the upper limit of normalization because when the current value equals the maximum value of the window, it has reached the highest level of this window. When the normalization result is between 0 and 0.3, it indicates that the current value is in the early part of the variation range of this window and is written as a low-amplitude range; when it is between 0.3 and 0.7, it indicates that the current value is in the middle part of the variation range of this window and is written as a medium-amplitude range; and when it is higher than 0.7 to 1, it indicates that the current value is in the later part of the variation range of this window and is written as a high-amplitude range.

[0028] The start and end times of the normalized temperature sequence of the cell tab, the normalized temperature sequence of the cell surface, the normalized pressure sequence of the battery surface, and the normalized strain sequence of the busbar are read respectively. The four start times are compared and the latest time is taken as the common starting point. The four end times are compared and the earliest time is taken as the common ending point. A common time axis is established between the common starting point and the common ending point according to a uniform sampling interval. Then, the four types of sequences are mapped to the common time axis one by one. When a measured value exists at the corresponding time, it is written directly. When one sampling interval is missing, the average of the measured value before and after the missing point is written. The single-point filling value = (measured value before the time + measured value after the time) / 2. When two to five consecutive sampling intervals are missing, the filling value is written for each missing time according to the time position ratio of the boundary values ​​before and after. The missing point filling value = front-end measured value + (time interval between the current missing time and the front-end time / total time interval between the back-end time and the front-end time) × (back-end measured value - front-end measured value). When more than five consecutive sampling intervals are missing, the segment is directly marked as an empty segment and no further splicing is performed. Subsequently, the normalized temperature values ​​of the cell tabs, the normalized temperature values ​​of the cell surface, the normalized pressure values ​​of the battery surface, and the normalized strain values ​​of the busbars at the same timestamp are written into the same row in a fixed column order, and then all time rows are stacked vertically in chronological order. When the number of null values ​​in the same row is equal to 0, it is written as a complete row; when the number of null values ​​is between 1 and 2, it is written as a usable row; when the number of null values ​​is greater than 2, it is written as an invalid row and deleted. After the above processing, a spatiotemporal sensing data matrix arranged according to a unified time base is obtained.

[0029] Step S202: Input the spatiotemporal sensing data matrix into the pre-trained deep learning model to obtain spatial coupling weights and temporal correlation weights, and perform weighted fusion calculation on the spatiotemporal sensing data matrix based on the spatial coupling weights and temporal correlation weights; The spatiotemporal sensing data matrix is ​​input into a pre-trained deep learning model (e.g., the Transformer model). The correlation distribution ratio between the value of each normalized sequence of the cell tab, cell surface, battery surface, and busbar and the value of each other normalized sequence at the same timestamp is calculated to obtain the spatial coupling weight. Simultaneously, the evolutionary distribution ratio between the value of each normalized sequence and the value at historical timestamps is calculated to obtain the temporal correlation weight. The specific process is as follows: The spatiotemporal sensing data matrix generated during historical operation is divided into multiple training samples with a fixed window length. Each training sample retains continuous records of four types of normalized data—cell tab temperature, cell surface temperature, battery surface pressure, and busbar strain—as they change over time. For each sample, the actual changes in each physical quantity over subsequent time periods are also recorded. The window length is determined by first setting multiple candidate lengths, forming samples for each, and then feeding them into the deep learning model one by one. The deviations of each candidate length from the subsequent changes on the validation samples are compared, and the length with the smallest deviation is selected as the final window length. Then, a parameter training process is established in a software program environment (Python environment). Each sample is written into a multi-row, four-column data table in chronological order, and a sequential position value is added to each row to distinguish the first, second, and final moments within the window. Next, a linear mapping is performed on each line of input values ​​and a bias value is added. Then, the current time line is matched with the other time lines in the same window one by one, and the absolute value of the difference between the two lines is calculated. The smaller the difference, the larger the temporary association value is written, and the larger the difference, the smaller the temporary association value is written. Then, all the temporary association values ​​corresponding to the same line are corrected to a set of proportions that sum to 1. The corrected proportions are used to generate the association allocation results in the training phase. The mapping parameters and bias parameters are repeatedly corrected by the deviation between the actual results of the training samples and the output results until the verification sample bias no longer decreases. After training, all mapping parameters and bias parameters are saved as a fixed parameter file. During the execution phase, this fixed parameter file is read, and the current spatiotemporal sensing data matrix is ​​input into the trained deep learning model. In the spatial direction, the normalized temperature values ​​of the cell tabs, the normalized temperature values ​​of the cell surface, the normalized pressure values ​​of the battery surface, and the normalized strain values ​​of the busbar at the same timestamp are paired one by one. The absolute value of the difference between the current item and another item is equal to |the normalized value of the current item - the normalized value of the other item|. Then, the absolute values ​​of the differences between the current item and the other three items are converted into temporary assigned values. The spatial coupling weight of a current item corresponding to a comparison item is calculated as the sum of the temporary assigned values ​​of the comparison item and the sum of the temporary assigned values ​​of all comparison items corresponding to the current item. In the temporal direction, the current value of a physical quantity is paired with all historical values ​​from the previous 1 time step, 2 time steps, and up to the end of the window. The absolute difference is calculated for each pair, then converted into temporary assigned values ​​and corrected to a proportional set summing to 1. The temporal correlation weight of the current value corresponding to a historical sampling time step is calculated as the sum of the temporary assigned values ​​of the historical sampling time step and the sum of the temporary assigned values ​​of all historical points corresponding to the current value. After the above processing, the spatial coupling weight and temporal correlation weight of the spatiotemporal sensing data matrix are obtained. When the spatial coupling weight or temporal correlation weight is between 0 and 0.2, it indicates that the item has the smallest proportion in the corresponding comparison set, written as a weak correlation interval. When it is between 0.2 and 0.6, it indicates that the item has a middle proportion in the corresponding comparison set, written as a medium correlation interval. When it is above 0.6 to 1, it indicates that the item has a dominant proportion in the corresponding comparison set, written as a strong correlation interval.

[0030] Based on spatial coupling weights and temporal correlation weights, a weighted summation calculation is performed on the spatiotemporal sensing data matrix to output the basic temperature feature sequences of the cell tabs, the basic temperature feature sequences of the cell surface, the basic pressure feature sequences of the battery surface, and the basic strain feature sequences of the busbar. These are then integrated into a weighted fused spatiotemporal sensing data matrix. The specific process is as follows: Read the normalized temperature value of the battery cell tab at the current moment and its spatial coupling weight from the spatiotemporal sensing data matrix, and read the normalized temperature values ​​of all battery cell tabs in the historical window and their corresponding temporal correlation weights. First, calculate the product of the current normalized temperature value of the battery cell tab and the spatial coupling weight, then calculate the product of the normalized temperature value of each battery cell tab in the historical window and the corresponding temporal correlation weight, and sum all the product results. The basic temperature feature value of the battery cell tab = current normalized temperature value of the battery cell tab × current spatial coupling weight of the battery cell tab + Σ(historical normalized temperature value of the battery cell tab × corresponding temporal correlation weight). The normalized temperature values ​​of all battery cell tabs within the historical window are derived from the previously constructed spatiotemporal sensing data matrix. Specifically, these are the values ​​of the corresponding normalized temperature columns of battery cell tabs in consecutive historical time rows extracted from the matrix before the current sampling time, according to a preset window length. The corresponding temporal correlation weights are derived from the deep learning model's calculation of the evolutionary correlation between the current normalized temperature value of the battery cell tabs and the normalized temperature values ​​of the battery cell tabs in the aforementioned historical time rows. Subsequently, following the same steps, the current normalized temperature value of the battery cell surface, the current spatial coupling weight of the battery cell surface, the historical normalized temperature value of the battery cell surface, and the corresponding temporal correlation weight are calculated. The basic temperature feature value of the battery cell surface is calculated as follows: current normalized temperature value of the battery cell surface × current spatial coupling weight of the battery cell surface + Σ(historical normalized temperature value of the battery cell surface × corresponding temporal correlation weight). Among them, the historical normalized temperature value of the cell surface comes from the same historical window before the current sampling time, that is, after truncating the preset window length from the spatiotemporal sensing data matrix according to the timestamp, the corresponding normalized temperature column of the cell surface is read; the corresponding temporal correlation weight comes from the temporal correlation allocation result of the deep learning model between the current normalized temperature value of the cell surface and the normalized temperature values ​​of each cell surface in the historical window. Then, following the same steps, read the current battery surface normalized pressure value, the current battery surface pressure spatial coupling weight, the historical battery surface normalized pressure value, and the corresponding temporal correlation weight. The battery surface basic pressure feature value = current battery surface normalized pressure value × current battery surface pressure spatial coupling weight + Σ(historical battery surface normalized pressure value × corresponding temporal correlation weight); where, the historical battery surface normalized pressure value comes from the corresponding battery surface normalized pressure column in the historical window before the current sampling time in the spatiotemporal sensing data matrix, representing the normalized change record of battery surface pressure at continuous historical sampling times; the corresponding temporal correlation weight comes from the deep learning model's evolutionary allocation ratio between the current battery surface normalized pressure value and the normalized pressure values ​​of each battery surface in the historical window. Then, following the same steps, read the current busbar normalized strain value, the current busbar spatial coupling weight, the historical busbar normalized strain value, and the corresponding time series correlation weight. The busbar basic strain characteristic value = current busbar normalized strain value × current busbar spatial coupling weight + Σ(historical busbar normalized strain value × corresponding time series correlation weight).The historical busbar normalized strain value is derived from the corresponding busbar normalized strain column within the historical window prior to the current sampling time in the spatiotemporal sensing data matrix, representing the normalized change record of busbar strain at consecutive historical sampling times. The corresponding temporal correlation weight is derived from the temporal correlation calculation results between the current busbar normalized strain value and the normalized strain values ​​of each busbar within the historical window calculated by the deep learning model. Finally, the cell tab base temperature feature value, cell surface base temperature feature value, battery surface base pressure feature value, and busbar base strain feature value at the same timestamp are written into the same row in a fixed column order and stacked vertically in chronological order to form a new matrix. After the above processing, a weighted fused spatiotemporal sensing data matrix is ​​obtained. In this matrix, any feature value between 0 and 0.25 indicates that the feature value is located at the lower end of the current window and is written as the low feature interval; between 0.25 and 0.75 indicates that the feature value is located in the middle of the current window and is written as the medium feature interval; and above 0.75 to 1 indicates that the feature value is located at the upper end of the current window and is written as the high feature interval.

[0031] Step S203: Based on the weighted fusion spatiotemporal sensing data matrix, determine the busbar strain response identifier, pressure change and heat accumulation abnormal coupling identifier, cell tab area temperature rise slope abnormal identifier, and cell surface temperature diffusion abnormal identifier, and align them according to the time stamp order to generate a battery thermal runaway feature sequence.

[0032] Based on the weighted fusion of spatiotemporal sensing data matrix, the rate of temperature change between adjacent time points in the basic temperature feature sequence of the battery cell tab is calculated. This rate is compared with a preset battery cell tab slope benchmark to obtain anomaly indicators for the temperature rise slope in the battery cell tab region. Simultaneously, the evolution and diffusion deviation of each temperature value within the basic temperature feature sequence of the battery cell surface is calculated and compared with a preset battery cell surface diffusion benchmark to obtain anomaly indicators for temperature diffusion on the battery cell surface. The specific process is as follows: Collect a spatiotemporal sensor data matrix that has undergone weighted fusion processing during historical normal operating periods. Calculate the rate of change of the basic temperature characteristic value of the battery cell tabs between adjacent moments, and summarize all rate of change results. Calculate the average and standard deviation of the historical battery cell tab temperature change rate. The lower limit of the battery cell tab slope benchmark = the average historical battery cell tab temperature change rate - the standard deviation of the historical battery cell tab temperature change rate; the upper limit of the battery cell tab slope benchmark = the average historical battery cell tab temperature change rate + the standard deviation of the historical battery cell tab temperature change rate. Then, calculate the battery cell surface temperature characteristic value point by point for the basic temperature characteristic value of the battery cell surface during historical normal operating periods. The surface evolution and diffusion deviation of the battery cell is calculated, and all deviation results are summarized to obtain the average value and standard deviation of the historical surface evolution and diffusion deviation of the battery cell. The lower limit of the battery cell surface diffusion reference = the average value of the historical surface evolution and diffusion deviation - the standard deviation of the historical surface evolution and diffusion deviation of the battery cell, and the upper limit of the battery cell surface diffusion reference = the average value of the historical surface evolution and diffusion deviation + the standard deviation of the historical surface evolution and diffusion deviation of the battery cell. Then, during the current operation, the basic temperature characteristic value of the battery cell electrode tab is read from the weighted fused spatiotemporal sensing data matrix time by time. The difference between the current time and the previous time is calculated, and then divided by the time interval between two adjacent time moments. Ear temperature change rate = (Current cell ear base temperature characteristic value - Previous cell ear base temperature characteristic value) / (Current time timestamp - Previous time timestamp); When the time interval is equal to 0, the point is deleted; When the cell ear temperature change rate is lower than the lower limit of the cell ear slope benchmark, it indicates that the current heating rate is lower than the lower boundary of the historical normal fluctuation zone, written as a low slope indicator; when the cell ear temperature change rate is between the lower limit and the upper limit of the cell ear slope benchmark, it indicates that the current heating rate is within the historical normal fluctuation zone, written as a normal slope indicator; when the cell ear temperature change rate is higher than the upper limit of the cell ear slope benchmark, it indicates that the current heating rate is within the historical normal fluctuation zone, written as a normal slope indicator; when the cell ear temperature change rate is higher than the upper limit of the cell ear slope benchmark, it indicates that the current heating rate is lower than the lower boundary of the historical normal fluctuation zone, written as a normal slope indicator. If the initial heating rate is higher than the upper boundary of the historical normal fluctuation zone, it is marked as an abnormal slope. Then, the basic temperature characteristic value of the cell surface is read, and the absolute difference between the current value and the previous value is calculated for each cell surface measuring point. Then, the absolute deviation between the current value of the measuring point and the average value of all cell surface measuring points at the same time is calculated. The average temperature characteristic value of the cell surface = the sum of the basic temperature characteristic values ​​of all cell surfaces at the same time / the number of cell surface measuring points. The cell surface evolution and diffusion deviation = |current cell surface basic temperature characteristic value - previous cell surface basic temperature characteristic value| + |current cell surface basic temperature characteristic value - average cell surface temperature characteristic value at the same time|.When the cell surface evolution diffusion deviation is below the lower limit of the cell surface diffusion reference, it indicates that both the temporal variation and spatial dispersion of the cell surface are below the lower boundary of the historical normal fluctuation zone, and is written as a diffusion contraction indicator. When the cell surface evolution diffusion deviation is between the lower and upper limits of the cell surface diffusion reference, it indicates that the cell surface change is within the historical normal fluctuation zone, and is written as a diffusion stability indicator. When the cell surface evolution diffusion deviation is above the upper limit of the cell surface diffusion reference, it indicates that the cell surface change is above the upper boundary of the historical normal fluctuation zone, and is written as a diffusion anomaly indicator. After the above processing, the abnormal temperature rise slope indicator and the abnormal temperature diffusion indicator of the cell surface are obtained.

[0033] Based on the weighted fusion of spatiotemporal sensing data matrix, the strain change deviation of each strain value in the busbar basic strain characteristic sequence is calculated. This deviation is compared with a preset busbar strain abrupt change benchmark to obtain the strain response abrupt change indicator for the busbar region. Simultaneously, the pressure change rate at adjacent time points in the battery surface basic pressure characteristic sequence is calculated, and combined with the temperature change rate, the coupling indicator between pressure change and abnormal heat accumulation is determined. The specific process is as follows: Collect a spatiotemporal sensor data matrix that has undergone weighted fusion processing during historical normal operation periods. Calculate the strain variation deviation of the busbar's basic strain characteristic values ​​point by point, and summarize all results to determine the historical average and standard deviation of strain variation deviations. The lower limit of the busbar strain mutation benchmark = historical average strain variation deviation - historical standard deviation of strain variation deviation; the upper limit of the busbar strain mutation benchmark = historical average strain variation deviation + historical standard deviation of strain variation deviation. Then, calculate the pressure change rate of the battery surface's basic pressure characteristic values ​​point by point, and summarize all results to determine the historical average and standard deviation of pressure change rate. The lower limit of the pressure change rate = historical average pressure change rate - historical standard deviation of pressure change rate. The standard deviation of the rate of change is calculated as follows: Upper limit of pressure rate of change = average historical pressure rate of change + standard deviation of historical pressure rate of change; then, the average and standard deviation of the historical cell tab temperature rate of change are calculated; Lower limit of temperature rate of change = average historical cell tab temperature rate of change - standard deviation of historical cell tab temperature rate of change; Upper limit of temperature rate of change = average historical cell tab temperature rate of change + standard deviation of historical cell tab temperature rate of change; then, during the current operation, the busbar foundation strain characteristic sequence is read, the absolute value of the difference between the strain value at the current moment and the previous moment is calculated, and the absolute value of the deviation between the strain value at the current moment and the average strain value of the historical window is calculated; the average strain value of the historical window = the strain value of all busbar foundations within the historical window. The sum of variable eigenvalues ​​ / number of historical window points, strain change deviation = |current busbar basic strain eigenvalue - previous busbar basic strain eigenvalue| + |current busbar basic strain eigenvalue - historical window strain average|; when the strain change deviation is lower than the lower limit of the busbar strain abrupt change benchmark, it is written as a strain smoothing indicator; when the strain change deviation is between the lower limit and the upper limit of the busbar strain abrupt change benchmark, it is written as a strain transition indicator; when the strain change deviation is higher than the upper limit of the busbar strain abrupt change benchmark, it is written as a strain abrupt change indicator; then read the battery surface basic pressure characteristic sequence, calculate the ratio of pressure difference between adjacent moments to time interval, pressure change rate = (current battery surface basic pressure eigenvalue - previous moment's data) / (current time's data) / (previous moment ... The formula is: (Basic pressure characteristic value of battery surface) / (Current time timestamp - Previous time timestamp); then the temperature change rate at the same time is called. When the pressure change rate is higher than the upper limit of the pressure change rate and the temperature change rate is higher than the upper limit of the temperature change rate, it is marked as a synchronous anomaly of pressure increase and heat accumulation; when the pressure change rate is higher than the upper limit of the pressure change rate and the temperature change rate is between the lower limit and the upper limit of the temperature change rate, it is marked as a pressure anomaly alone; when the pressure change rate is between the lower limit and the upper limit of the pressure change rate and the temperature change rate is higher than the upper limit of the temperature change rate, it is marked as a heat accumulation anomaly alone; when the pressure change rate is lower than the lower limit of the pressure change rate and the temperature change rate is higher than the upper limit of the temperature change rate, it is marked as a pressure drop accompanied by heat anomaly.When the pressure change rate is between its lower and upper limits, and the temperature change rate is between its lower and upper limits, it is written as a normal mechanical-thermal consistency indicator. After the above processing, the manifold strain response indicator and the abnormal coupling indicator of pressure change and heat accumulation are obtained.

[0034] Finally, the busbar strain response identifier, pressure change and heat accumulation abnormal coupling identifier, cell tab area temperature rise slope abnormal identifier, and cell surface temperature diffusion abnormal identifier are written into the sequence according to a unified timestamp and a fixed order to obtain the battery thermal runaway characteristic sequence.

[0035] Step S300: Based on the battery thermal runaway feature sequence and the pre-fitted feature decoding parameter set, determine the hot spot temperature prediction data and the remaining thermal runaway time prediction data.

[0036] Specifically, such as Figure 3 As shown, based on the battery thermal runaway characteristic sequence and the pre-fitted feature decoding parameter set, hotspot temperature prediction data and thermal runaway remaining time prediction data are determined, including: Step S301: Extract the decoding weight matrix and decoding bias vector from the pre-fitted feature decoding parameter set, and merge the decoding weight matrix and decoding bias vector with the battery thermal runaway feature sequence to obtain the hidden feature vector.

[0037] Based on a pre-fitted feature decoding parameter set constructed from historical battery multiphysics runaway experimental data, the decoding weight matrix and decoding bias vector are extracted and merged with the battery thermal runaway feature sequence to obtain the latent feature vector. The specific process is as follows: Historical operation records and multiphysics runaway experimental data records generated under controlled conditions were collected, retaining cell tab temperature, surface temperature, surface pressure, bus strain, and corresponding subsequent hot spot temperature and remaining time results. These records were then subjected to normalization, spatiotemporal alignment, correlation weight calculation, weighted feature output, temperature rise diffusion comparison, and mechanical-thermal coupling judgment to obtain historical battery thermal runaway feature sequences. Each historical battery thermal runaway feature sequence was paired with its subsequent real hot spot temperature and real remaining time to form training samples. Each position in the historical battery thermal runaway feature sequence was then numerically quantified: a normal indicator was written as 0, indicating that the position had not deviated from the normal fluctuation zone; a transition indicator was written as 0.5, indicating that the position had deviated from the normal center but had not crossed the abnormal boundary; an abnormal indicator was written as 1, indicating that the position had crossed the abnormal boundary; and an empty state was written as -1, indicating that the position was missing and needed to be distinguished from the other three states. Finally, all training samples were input into the decoding parameter fitting training process, which was used to determine the values ​​of the decoding weight matrix and decoding bias vector based on the historical samples. Specifically, the quantified historical battery thermal runaway feature sequence is used as the input variable, and the subsequent real hotspot temperature and real remaining time corresponding to the historical sample are used as supervision labels. First, the correspondence between the input variable and the supervision label is established according to the sample number and timestamp. Then, the input variable is multiplied and accumulated item by item with the current decoding weight matrix, and the decoding bias vector is superimposed to obtain the corresponding latent feature expression result. Subsequently, based on the prediction deviation between the latent feature expression result and the subsequent real hotspot temperature and real remaining time, the values ​​of the decoding weight matrix and the decoding bias vector are adjusted in reverse, so that the same type of thermal runaway symptoms gradually gather in the latent feature space, and the latent features corresponding to different risk states are expressed in different ways. Features gradually become more differentiated. After multiple iterations, when the prediction bias of the validation samples no longer decreases or the decrease is below the preset convergence threshold, parameter updates are stopped, and the final decoded weight matrix and decoded bias vector are saved as a feature decoding parameter set. During the execution phase, the feature decoding parameter set is read, and the battery thermal runaway feature sequence under the current timestamp is numerically converted and mapped row by row to the decoded weight matrix. The i-th hidden component = Σ(the decoded weight value of the i-th row × the feature fingerprint value at the corresponding position) + the decoded bias value of the i-th row; where i is the component index of the hidden feature vector, i = 1, 2, ..., m, and m is the number of rows in the decoded weight matrix, which is also the number of dimensions of the hidden feature vector. After completing all row operations in sequence, the hidden components of each row are arranged in a fixed order to form the hidden feature vector. After the above processing, the hidden feature vector is obtained. When the hidden component is less than 0, it means that the component is offset in the opposite direction relative to the historical center value, which is written as the reverse deviation interval. When it is between 0 and 1, it means that the component is offset in a small way, which is written as the low response interval. When it is between 1 and 3, it means that the component is offset in the middle, which is written as the medium response interval. When it is higher than 3, it means that the component has moved significantly away from the historical center zone, which is written as the high response interval.

[0038] Step S302: Extract the temperature prediction regression vector and the time prediction regression vector from the pre-fitted feature decoding parameter set, calculate the inner product value between the hidden feature vector and the temperature prediction regression vector to obtain the hotspot temperature prediction data, and calculate the inner product value between the hidden feature vector and the time prediction regression vector to obtain the thermal runaway remaining time prediction data.

[0039] The fingerprint latent feature vector, actual hotspot temperature, and actual remaining time from historical samples are written into the training and validation sets, respectively. A dual-output linear regression model is used as the prediction regression model, which includes a temperature prediction regression branch and a remaining time prediction regression branch. The temperature prediction regression branch is used to learn the mapping relationship between the latent feature vector and the predicted hotspot temperature data, and the remaining time prediction regression branch is used to learn the mapping relationship between the latent feature vector and the predicted remaining time of thermal runaway. Coefficients are corrected round by round on the training set to simultaneously reduce the deviation between the predicted hotspot temperature and the actual hotspot temperature, and the deviation between the predicted remaining time and the actual remaining time. Correction stops when the deviation in the validation set no longer decreases after several rounds, and the final temperature prediction regression vector, time prediction regression vector, temperature prediction bias value, and time prediction bias value are saved. During the execution phase, the saved regression vector and bias value are read, the fingerprint latent feature vector at the current moment is matched with the temperature prediction regression vector bit by bit and summed, and then the temperature prediction bias value is added. Hotspot temperature prediction data = Σ(fingerprint latent feature component value × corresponding temperature) The predicted regression coefficient is calculated by adding the temperature prediction bias value. Then, the same fingerprint latent feature vector is matched bit-by-bit with the time prediction regression vector and summed, then the time prediction bias value is added. The predicted remaining time of thermal runaway is calculated as: Σ(fingerprint latent feature component value × corresponding time prediction regression coefficient) + time prediction bias value. After the above processing, the hotspot temperature prediction data and the predicted remaining time of thermal runaway are obtained. When the predicted hotspot temperature is below 45 degrees Celsius, it indicates that the predicted hotspot is still in the normal operating and slightly warming range, written as the low temperature prediction interval, which is within 45 degrees Celsius. When the temperature reaches 60 degrees Celsius, it indicates that the predicted hotspot has entered a significant warming range, written as the medium-temperature prediction range. When the temperature is above 60 degrees Celsius, it indicates that the predicted hotspot has entered a higher temperature range, written as the high-temperature prediction range. When the remaining time prediction data is above 1800 seconds, it indicates that the subsequent event is still a long time away from the current moment, written as the long remaining time range. When it is between 300 and 1800 seconds, it indicates that the subsequent event is a medium time away from the current moment, written as the medium remaining time range. When it is below 300 seconds, it indicates that the subsequent event is very short away from the current moment, written as the short remaining time range.

[0040] Step S400: The hotspot temperature prediction data and the thermal runaway remaining time prediction data are calibrated according to the historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set to obtain the upper confidence boundary set, the lower confidence boundary set and the interval width set.

[0041] Specifically, such as Figure 4 As shown, the hotspot temperature prediction data and the thermal runaway remaining time prediction data are calibrated based on the historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set, resulting in the upper confidence boundary set, the lower confidence boundary set, and the interval width set, including: Step S401: Arrange the historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set in descending order. According to the preset distribution quantile, extract the corresponding historical error quantile values ​​from the historical temperature calibration set and the historical remaining time calibration set after descending order as compensation range parameters, and establish the hotspot temperature prediction calibration margin and the thermal runaway remaining time prediction calibration margin.

[0042] The process involves acquiring the historical cumulative measured sequence of battery hotspot temperatures and the preset baseline sequence of battery hotspot temperatures, as well as the historical measured sequence of remaining battery thermal runaway time and the preset baseline sequence of remaining battery thermal runaway time. The deviation between the historical measured sequence of cumulative battery temperatures and the preset baseline sequence of battery temperatures is calculated to generate a historical hotspot temperature calibration set. Finally, the deviation between the historical measured sequence of remaining battery thermal runaway time and the preset baseline sequence of remaining battery thermal runaway time is calculated to obtain the historical remaining thermal runaway time calibration set. The specific process is as follows: Historical operational cumulative data is collected, which is continuously acquired during the battery's historical charge-discharge operation and historical multiphysics runaway experiments. The recorded content includes at least the sampling timestamp, the actual measured battery temperature, the time of thermal runaway occurrence or the marked end point of thermal runaway, the charge rate, the discharge rate, the ambient temperature, and the initial state of charge. The battery's historical operational cumulative temperature measurement sequence is obtained by extracting the actual measured battery temperature values ​​at each sampling time from the historical operational cumulative data, and then cleaning, aligning, and arranging them according to the timestamps. The battery thermal runaway remaining time measurement sequence... By reading historical samples with thermal runaway occurrence times or thermal runaway endpoints, the remaining time is measured using the time difference between each sampling time and the corresponding thermal runaway occurrence time or thermal runaway endpoint, and then arranged in timestamp order. The actual measured battery temperature, remaining time, and corresponding charge rate, discharge rate, ambient temperature, and initial state of charge are retained. The historical samples are then grouped according to charge rate, discharge rate, ambient temperature, and initial state of charge. The preset battery temperature reference sequence is obtained by averaging the temperature values ​​at the same relative time position within the same operating condition group. The preset reference sequence for the remaining time of battery thermal runaway is obtained by averaging the remaining time values ​​at the same relative time position within the same operating condition group. The preset reference sequence for battery temperature is formed by averaging the temperature values ​​at the same relative time position within the same group. The preset reference sequence for the remaining time within the same group is then formed by averaging the corresponding remaining times. Next, the historical cumulative temperature measurement sequence of the battery and the preset reference sequence for battery temperature are paired point-by-point at the same sampling time. The single-point temperature calibration error is calculated as |measured temperature value at the corresponding time - preset reference temperature value at the corresponding time|. Finally, the battery thermal runaway... The measured remaining time sequence and the preset baseline sequence of remaining time for battery thermal runaway are paired point by point at the same sampling time. The single-point remaining time calibration error = |measured remaining time value at the corresponding time - preset baseline value of remaining time at the corresponding time|. When the timestamps of any pairing point are inconsistent, the nearest neighbor point with a time difference of no more than 1 sampling period is selected for pairing. If the time difference is greater than 1 sampling period, the pairing point is deleted. All single-point temperature calibration errors are written into the historical hot spot temperature calibration set in chronological order, and all single-point remaining time calibration errors are written into the historical thermal runaway remaining time calibration set in chronological order. After the above processing, we obtain the historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set. Among them, when the temperature calibration error is between 0 degrees Celsius and 2 degrees Celsius, it means that the difference between the actual temperature and the reference temperature under the same operating condition is still within the common fluctuation range, which is written as the small deviation range. When it is between 2 degrees Celsius and 5 degrees Celsius, it means that the deviation is higher than the common fluctuation but not significantly amplified, which is written as the medium deviation range. When it is higher than 5 degrees Celsius, it means that the temperature deviation has significantly increased, which is written as the large deviation range. When the remaining time calibration error is between 0 seconds and 60 seconds, it is written as the small deviation range. When it is between 60 seconds and 180 seconds, it is written as the medium deviation range. When it is higher than 180 seconds, it is written as the large deviation range.

[0043] The historical hotspot temperature calibration set is sorted from largest to smallest. Multiple candidate distribution quantiles are then defined. For each candidate distribution quantile, the error value corresponding to its sorting position is extracted as a compensation amount. This compensation amount is added to the historical predicted values, and the coverage ratio of the actual values ​​falling within the compensation boundary is calculated. The boundary width is also recorded. The candidate distribution quantile with the narrowest boundary width and the coverage ratio meeting the target requirement is selected as the preset distribution quantile. Subsequently, during the execution phase, the corresponding error values ​​are extracted from the sorted historical hotspot temperature calibration set according to the preset distribution quantile. If the preset distribution quantile is... If the quantile is 95%, then the error value at the top 5% boundary of the total sample size is taken. If the preset distribution quantile is 90%, then the error value at the top 10% boundary of the total sample size is taken. If the preset distribution quantile is other values, then the error value at the corresponding proportion position is taken. The sorting position = total sample size × (1 - preset quantile). Then write this error value as the temperature prediction calibration margin. Then, follow the same steps to sort the historical thermal runaway remaining time calibration set from largest to smallest, and extract the corresponding error value according to the preset distribution quantile, and write it as the remaining time prediction calibration margin. After the above processing, the hotspot temperature prediction calibration margin and the remaining time prediction calibration margin are obtained. Among them, when the temperature prediction calibration margin is between 0 degrees Celsius and 2 degrees Celsius, it indicates that the historical error distribution is concentrated and the boundary expansion is small, which is written as a narrow margin interval. When it is between 2 degrees Celsius and 5 degrees Celsius, it indicates that the historical error has moderate dispersion, which is written as a medium margin interval. When it is above 5 degrees Celsius, it indicates that the historical error tail is long and the boundary needs to be significantly expanded, which is written as a wide margin interval. When the remaining time prediction calibration margin is between 0 seconds and 60 seconds, it is written as a narrow margin interval. When it is between 60 seconds and 180 seconds, it is written as a medium margin interval. When it is above 180 seconds, it is written as a wide margin interval.

[0044] Step S402: Calculate the summative and subtractive extreme values ​​of the hotspot temperature prediction data and the thermal runaway remaining time prediction data with the corresponding hotspot temperature prediction calibration margin and thermal runaway remaining time prediction calibration margin, respectively. Merge each summative extreme value to obtain the upper confidence boundary set and merge each subtractive extreme value to obtain the lower confidence boundary set. Calculate the extreme value deviation between the corresponding values ​​of the boundary sets to obtain the interval width set.

[0045] By mapping the hotspot temperature prediction data to the temperature prediction calibration margin item by item, the upper and lower temperature boundary values ​​are calculated. The upper temperature boundary value = hotspot temperature prediction data + temperature prediction calibration margin, and the lower temperature boundary value = hotspot temperature prediction data - temperature prediction calibration margin. Then, by mapping the thermal runaway remaining time prediction data to the remaining time prediction calibration margin item by item, the upper and lower time boundary values ​​are calculated. The upper time boundary value = thermal runaway remaining time prediction data + remaining time prediction calibration margin, and the lower time boundary value = thermal runaway remaining time prediction data - remaining time prediction margin. Measure the calibration margin; then write all upper temperature and upper time boundary values ​​into the upper confidence boundary set, and write all lower temperature and lower time boundary values ​​into the lower confidence boundary set; then calculate the absolute difference between the upper and lower boundary values ​​of the same index, and the extreme value deviation is the absolute difference between the upper and lower boundary values ​​of the same index. The temperature interval width = |temperature upper boundary value - lower temperature lower boundary value|, and the time interval width = |time upper boundary value - time lower boundary value|; write all temperature interval widths and time interval widths into the interval width set in chronological order. After the above processing, we obtain the upper confidence boundary set, the lower confidence boundary set, and the interval width set. For example, at a certain sampling time, the hotspot temperature prediction data is 58 degrees Celsius, and the temperature prediction calibration margin is 3 degrees Celsius. Then, the upper temperature boundary value is 61 degrees Celsius, the lower temperature boundary value is 55 degrees Celsius, and the temperature interval width is 6 degrees Celsius. At the same sampling time, the thermal runaway remaining time prediction data is 500 seconds, and the remaining time prediction calibration margin is 80 seconds. Then, the upper time boundary value is 580 seconds, the lower time boundary value is 420 seconds, and the time interval width is 160 seconds. The upper temperature boundary value and the upper time boundary value corresponding to this sampling time are written into the upper confidence boundary set, the lower temperature boundary value and the lower time boundary value are written into the lower confidence boundary set, and the temperature interval width and the time interval width are written into the interval width set. After the above processing, we obtain the upper confidence boundary set, the lower confidence boundary set, and the interval width set. When the interval width is between 0 and 2 degrees Celsius or between 0 and 60 seconds, it indicates that the boundary expansion range is relatively narrow and is written as a narrow interval. When it is between 2 and 5 degrees Celsius or between 60 and 180 seconds, it indicates that the boundary expansion range is in the middle and is written as a medium interval. When it is higher than 5 degrees Celsius or 180 seconds, it indicates that the boundary expansion range is significantly increased and is written as a wide interval.

[0046] Step S500: Generate battery thermal runaway prediction results based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set; and output battery thermal management control commands based on the battery thermal runaway prediction results.

[0047] Specifically, battery thermal runaway prediction results are generated based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set, including: The battery thermal runaway prediction result is obtained by comparing the summation distribution extreme value of each item in the upper confidence boundary set with a preset temperature risk judgment threshold, comparing the subtraction distribution extreme value of each item in the lower confidence boundary set with a preset time emergency judgment threshold, and comparing the interval width set with a preset anomaly amplification width threshold. The specific process is as follows: Historical samples with actual subsequent state records are collected. These records refer to the actual operational status of the historical sample within the subsequent observation window, including whether hotspot temperatures exceeded limits, thermal runaway occurred, the critical temperature rise state was reached, abnormal pressure increases occurred, abnormal busbar strain occurred, and whether the system recovered to normal after cooling or flow restriction. For each sample, the upper temperature boundary value, lower time boundary value, interval width, and corresponding actual subsequent state are retained. The samples are then divided into normal, critical, and high-risk groups based on their actual subsequent states. Subsequently, the upper temperature boundary position that separates the critical group from the normal group is identified in the historical samples, and this position is designated as the preset temperature risk. The system first determines the threshold value; then it finds the lower time boundary position that can separate emergency state samples from non-emergency state samples, and writes this position as the preset time emergency determination threshold; then it finds the width position that can separate boundary stable samples from boundary amplified samples, and writes this position as the abnormal amplification width threshold; then, during the current operation, it reads the upper temperature boundary value item by item from the upper confidence boundary set and compares it with the preset temperature risk determination threshold. When the upper temperature boundary value is higher than the preset temperature risk determination threshold, it is written as a temperature exceeding the limit state; when the upper temperature boundary value is equal to the preset temperature risk determination threshold, it is written as a temperature critical state; when the upper temperature boundary value is lower than the preset temperature risk determination threshold, it is written as a temperature not exceeding the limit state; and so on. Next, the lower confidence boundary values ​​are read item by item from the lower confidence boundary set and compared with a preset time emergency judgment threshold. When the lower confidence boundary value is lower than the preset time emergency judgment threshold, it is written as a time emergency state; when the lower confidence boundary value is equal to the preset time emergency judgment threshold, it is written as a time critical state; and when the lower confidence boundary value is higher than the preset time emergency judgment threshold, it is written as a time non-emergency state. Then, the interval widths are read item by item from the interval width set and compared with an abnormal amplification width threshold. When the interval width is higher than the abnormal amplification width threshold, it is written as a width amplification state; when the interval width is equal to the abnormal amplification width threshold, it is written as a width critical state; and when the interval width is lower than the abnormal amplification width threshold, it is written as a width stable state. The system then processes the following states: temperature, time, and width, combined using the same timestamp. If a temperature exceedance, time emergency, and width amplification state occur simultaneously, a high-risk prediction result is output. If these conditions occur simultaneously, an emergency risk prediction result is output. If a temperature exceedance and time non-emergency state occurs, a heat accumulation risk prediction result is output. If a temperature does not exceed the limit and time emergency state occurs, a sudden risk prediction result is output. If any of the temperature, time, or width critical states occur, a critical observation prediction result is output. If a temperature does not exceed the limit, time non-emergency state, and width stable state occur simultaneously, a normal prediction result is output. After this processing, the battery thermal runaway prediction results are obtained as follows: high risk, emergency risk, heat accumulation risk, sudden risk, critical observation, and normal.

[0048] Specifically, based on the battery thermal runaway prediction results, battery thermal management control commands are output, including: The system outputs battery thermal management control commands for targeted physical cooling of battery risk areas indicated by battery thermal runaway prediction results, and / or outputs battery thermal management control commands for regulating current in the corresponding current transmission paths of battery risk areas indicated by battery thermal runaway prediction results. The specific process is as follows: Based on the battery thermal runaway prediction results, the coordinates of the corresponding risk location and the current transmission path number are read, and the execution parameter table of the thermal management equipment is called. The execution parameter table is formed by recording the stable output levels after the thermal management equipment of the same type has been adjusted step by step within the rated flow and rated current range. The thermal management equipment includes one or more of the following: liquid cooling plate, cooling pipeline, micro pump, proportional valve, air-cooled fan, heat-conducting structural components, power switch, relay or contactor; the execution end refers to the controlled components such as valve, pump, fan, power switch, relay or contactor that receive controller instructions and perform cooling flow regulation, current limiting, current cut-off or path switching actions. When the prediction result is high-risk, first issue a power switch disconnection command (current cut-off) to the corresponding path, then issue a cooling command of 80% to 100% of the rated flow to the corresponding risk area; when the prediction result is emergency risk, first issue a current limiting command to the corresponding path, reducing the current to 20% to 40% of the rated current, then issue a cooling command of 50% to 80% of the rated flow to the corresponding risk area; when the prediction result is heat accumulation risk, keep the path open and issue a cooling command of 30% to 60% of the rated flow to the corresponding risk area; when the prediction result is sudden risk, first issue a current interruption command to the corresponding path, then issue a fixed-point cooling command of 60% to 90% of the rated flow to the adjacent area of ​​the busbar; when the prediction result is critical observation prediction, maintain the original current transmission state and issue a cooling command of 10% to 30% of the rated flow. The system executes the cooling command. When the prediction result is normal, it maintains the original path state and keeps the cooling flow rate between 0% and 10% of the rated flow rate. Then, the execution end returns the valve position opening, switch status, flow feedback value, and execution time, and compares each item with the target command. Valve position deviation = |feedback valve position opening - target valve position opening|. When the valve position deviation is between 0% and 5%, it is written as "execution in place". When the valve position deviation is between 5% and 10%, it is written as "execution lag". When the valve position deviation is higher than 10%, it is written as "execution deviation". Then, the current switch status is compared with the target status. If they are consistent, it is written as "path action completed". If they are inconsistent, it is written as "path action not completed". Then, the flow feedback value is compared with the target flow rate. Flow deviation = |feedback flow rate value - target flow rate value| / target flow rate value. When the flow deviation is between 0% and 10%, it is written as "flow effective". When the flow deviation is higher than 10%, it is written as "flow mismatch".

[0049] The battery thermal management method in this embodiment of the invention has at least the following beneficial effects: (1) This invention deploys distributed temperature sensing, distributed pressure sensing, and distributed strain sensing on a flexible printed circuit board, and directly attaches the flexible printed circuit board to the monitoring structure of the battery cell tabs, battery cell surface, and key busbar locations. Instead of setting up a single sensor, it places the three types of sensing objects—heat, force, and deformation—into the same attachment carrier and the same continuous acquisition link, so that the local temperature rise, surface pressure, and busbar deformation of the battery during charging and discharging can be recorded synchronously at the same time coordinate. Taking the scenario where the battery cell tabs first show local heating, then the battery surface pressure changes, and strain fluctuations occur in the busbar connection area as an example, the corresponding records of multiple key points within the same time period can be directly obtained, rather than just the discrete results of a single temperature point. That is, by capturing early signs through multi-physics coupling characteristics, it has the ability to provide early warning of thermal runaway, and improves the credibility of the alarm based on the confidence boundary, solving the technical problems of low credibility of thermal runaway alarms and delayed thermal runaway warnings caused by alarms based on preset temperature thresholds in the prior art.

[0050] (2) This invention constructs a unified spatiotemporal sensing data matrix and introduces a fusion processing method that combines spatial coupling weights and temporal correlation weights on this matrix. It does not simply put multiple types of data together, but first normalizes and aligns the cell tab temperature, cell surface temperature, battery surface pressure, and busbar strain in time. Then, it performs weighted output based on the correspondence between different physical quantities at the same moment and the evolution relationship between the current moment and historical moments, forming a feature sequence of basic temperature, basic pressure, and basic strain. This advances the collection of thermal information and structural response information from parallel acquisition to correlation and fusion. For example, when the temperature rise of the cell tab is not large but the busbar strain and surface pressure have shifted synchronously, this processing method can still reflect the abnormal trend through the weighted feature relationship, rather than relying solely on the absolute magnitude of a single measurement.

[0051] (3) This invention achieves early identification of signs of thermal runaway through a continuous process of temperature rise and diffusion comparison, force-thermal coupling judgment, feature fingerprint generation, implicit parameter mapping, and joint prediction of hot spot temperature and remaining time. The innovation here is that it does not simply give a temperature alarm value, but simultaneously generates multiple types of indicators such as abnormal temperature rise slope of cell tabs, abnormal surface diffusion, sudden change in busbar strain, and abnormal coupling of pressure and heat accumulation. These indicators are then combined into a feature fingerprint sequence, which is further mapped to obtain early hot spot temperature prediction data and thermal runaway remaining time prediction data. Taking a scenario where the overall surface temperature of a battery has not yet increased significantly, but the local temperature rise rate of the cell tabs has accelerated and the busbar strain response has changed abruptly, this link can give trend judgment at an earlier stage.

[0052] (4) This invention employs a boundary division method using a set of historical calibration errors, a prediction margin, and upper and lower confidence boundaries. It utilizes historical temperature errors and historical remaining time errors to form a compensation margin, and then expands the boundaries of the hotspot temperature prediction results and the remaining time prediction results to obtain a set of interval widths. Based on this, the risk level is determined. In this way, the predicted value and the historical error distribution are incorporated into the judgment logic, so that the risk output is no longer a single-point judgment, but a graded result with boundary widths and confidence ranges. For example, when the predicted hotspot temperatures of two battery samples are close, if the interval width of one sample is significantly larger, it can be classified as a state that requires more attention.

[0053] (5) This invention directly links the aforementioned risk results to the intervention targets and current transmission path control methods of the thermal management equipment. Based on the risk location, risk level, and corresponding path relationship, it implements targeted cooling physical intervention or current adjustment, making monitoring, fusion, prediction, boundary division, and control form a closed execution link. Taking the scenario where the area adjacent to the busbar is determined to be high-risk and the remaining time enters the emergency zone as an example, it can directly correspond to the area and its connected path to implement current interruption and fixed-point cooling, rather than treating the entire battery with a uniform intensity. This establishes a clear mapping relationship between the risk identification results and the intervention targets, and the control actions are targeted at the location and the path.

[0054] like Figure 5 As shown in the figure, this embodiment of the invention also provides a battery thermal management device, including: an acquisition unit, a construction unit, a determination unit, a calibration unit, and an output unit.

[0055] The acquisition unit is used to acquire monitoring data of key parts of the battery, including the monitoring temperature of the cell tabs, the monitoring temperature of the cell surface, the monitoring pressure of the battery surface, and the monitoring strain of the busbar.

[0056] The construction unit is used to construct a spatiotemporal sensing data matrix based on the monitoring data, and to generate a battery thermal runaway feature sequence based on the spatiotemporal sensing data matrix.

[0057] The determining unit is used to determine hotspot temperature prediction data and thermal runaway remaining time prediction data based on the battery thermal runaway feature sequence and the pre-fitted feature decoding parameter set.

[0058] The calibration unit is used to calibrate the hotspot temperature prediction data and the thermal runaway remaining time prediction data according to the historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set, to obtain the upper confidence boundary set, the lower confidence boundary set and the interval width set.

[0059] The output unit is used to generate battery thermal runaway prediction results based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set, and to output battery thermal management control commands based on the battery thermal runaway prediction results.

[0060] This invention also provides a computer device, including: a memory, a processor, and a network interface connected via a system bus, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement all or part of the steps of the aforementioned battery thermal management method.

[0061] The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0062] A processor can be a CPU, or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor, or any conventional processor. The processor is the control center of a computer device, connecting all parts of the computer device through various interfaces and lines.

[0063] Memory can be used to store computer programs and / or modules. The processor implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and by accessing data stored in the memory. Memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, application programs required for at least one function (such as video playback, image playback, etc.), etc.; the data storage area can store data created using the memory (such as video data, image data, etc.). Furthermore, memory can include high-speed random access memory (RAM), and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital cards (SD cards), flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0064] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements all or part of the steps of the aforementioned battery thermal management method.

[0065] The embodiments of the present invention can implement all or part of the aforementioned processes, or they can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various methods described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0066] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, servers, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0067] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0068] The serial numbers in the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0069] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A battery thermal management method, characterized in that, Includes the following steps: Acquire monitoring data for key components of the battery, including cell tab monitoring temperature, cell surface monitoring temperature, battery surface monitoring pressure, and busbar monitoring strain; A spatiotemporal sensing data matrix is ​​constructed based on the monitoring data, and a battery thermal runaway characteristic sequence is generated based on the spatiotemporal sensing data matrix. Based on the battery thermal runaway feature sequence and the pre-fitted feature decoding parameter set, determine the hot spot temperature prediction data and the thermal runaway remaining time prediction data; The hotspot temperature prediction data and the thermal runaway remaining time prediction data are calibrated based on the historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set to obtain the upper confidence boundary set, the lower confidence boundary set and the interval width set. The battery thermal runaway prediction results are generated based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set. The battery thermal management control command is then output based on the battery thermal runaway prediction results.

2. The battery thermal management method according to claim 1, characterized in that, The step of constructing a spatiotemporal sensing data matrix based on the monitoring data and generating a battery thermal runaway feature sequence based on the spatiotemporal sensing data matrix includes: The monitoring data is numerically normalized, and the normalized monitoring data is aligned according to the time stamp order to construct a spatiotemporal sensing data matrix. The spatiotemporal sensing data matrix is ​​input into a pre-trained deep learning model to obtain spatial coupling weights and temporal correlation weights. The spatiotemporal sensing data matrix is ​​then weighted and fused based on the spatial coupling weights and temporal correlation weights. Based on the weighted fusion spatiotemporal sensing data matrix, the busbar strain response identifier, the pressure change and heat accumulation abnormal coupling identifier, the cell tab area temperature rise slope abnormal identifier, and the cell surface temperature diffusion abnormal identifier are determined, and they are aligned in chronological order according to the timestamps to generate the battery thermal runaway feature sequence.

3. The battery thermal management method according to claim 1, characterized in that, The step of determining hotspot temperature prediction data and thermal runaway remaining time prediction data based on the battery thermal runaway feature sequence and a pre-fitted feature decoding parameter set includes: Extract the decoding weight matrix and decoding bias vector from the pre-fitted feature decoding parameter set, and merge the decoding weight matrix and decoding bias vector with the battery thermal runaway feature sequence to obtain the hidden feature vector; Temperature prediction regression vector and time prediction regression vector are extracted from the pre-fitted feature decoding parameter set. The inner product value between the hidden feature vector and the temperature prediction regression vector is calculated to obtain hotspot temperature prediction data. The inner product value between the hidden feature vector and the time prediction regression vector is also calculated to obtain thermal runaway remaining time prediction data.

4. The battery thermal management method according to claim 1, characterized in that, The process involves calibrating the predicted hotspot temperature and predicted thermal runaway time based on historical hotspot temperature calibration sets and historical thermal runaway time calibration sets to obtain an upper confidence boundary set, a lower confidence boundary set, and an interval width set, including: The historical hotspot temperature calibration set and the historical thermal runaway remaining time calibration set are arranged in descending order. Based on the preset distribution quantile, the corresponding historical error quantile values ​​are extracted from the historical temperature calibration set and the historical remaining time calibration set after descending order as compensation range parameters to establish the hotspot temperature prediction calibration margin and the thermal runaway remaining time prediction calibration margin. Calculate the summative and subtractive extreme values ​​of the hotspot temperature prediction data and the thermal runaway remaining time prediction data with the corresponding hotspot temperature prediction calibration margin and thermal runaway remaining time prediction calibration margin, respectively. Merge each summative extreme value to obtain the upper confidence boundary set and each subtractive extreme value to obtain the lower confidence boundary set. Calculate the extreme value deviation between the corresponding values ​​of the boundary sets to obtain the interval width set.

5. The battery thermal management method according to claim 4, characterized in that, The process of generating battery thermal runaway prediction results based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set includes: The extreme values ​​of the summation distribution of each item in the upper confidence boundary set are compared with the preset temperature risk judgment threshold, the extreme values ​​of the subtraction distribution of each item in the lower confidence boundary set are compared with the preset time emergency judgment threshold, and the interval width set is compared with the preset anomaly amplification width threshold to obtain the battery thermal runaway prediction result.

6. The battery thermal management method according to claim 1, characterized in that, The step of outputting battery thermal management control commands based on battery thermal runaway prediction results includes: The system outputs a battery thermal management control command to perform directional physical cooling on the battery risk sites indicated by the battery thermal runaway prediction results, and / or outputs a battery thermal management control command to regulate the current in the current transmission path corresponding to the battery risk sites indicated by the battery thermal runaway prediction results.

7. The battery thermal management method according to claim 1, characterized in that, The acquisition of monitoring data for key battery components includes: By attaching a flexible printed circuit board to key parts of the battery, the induced electrical signals from temperature sensors, pressure sensors, and strain sensors integrated on the flexible printed circuit board are acquired, and monitoring data of key parts of the battery are obtained based on the induced electrical signals.

8. A battery thermal management device, characterized in that, include: The acquisition unit is used to acquire monitoring data of key parts of the battery, including the monitoring temperature of the cell tabs, the monitoring temperature of the cell surface, the monitoring pressure of the battery surface, and the monitoring strain of the busbar. The construction unit is used to construct a spatiotemporal sensing data matrix based on the monitoring data, and generate a battery thermal runaway feature sequence based on the spatiotemporal sensing data matrix. The determining unit is used to determine the hot spot temperature prediction data and the thermal runaway remaining time prediction data based on the battery thermal runaway feature sequence and the pre-fitted feature decoding parameter set. The calibration unit is used to calibrate the hot spot temperature prediction data and the thermal runaway remaining time prediction data according to the historical hot spot temperature calibration set and the historical thermal runaway remaining time calibration set, to obtain the upper confidence boundary set, the lower confidence boundary set and the interval width set. The output unit is used to generate battery thermal runaway prediction results based on the upper confidence boundary set, the lower confidence boundary set, and the interval width set, and to output battery thermal management control commands based on the battery thermal runaway prediction results.

9. A computer device, characterized in that, include: A memory and a processor, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the battery thermal management method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the battery thermal management method according to any one of claims 1 to 7.

Images