Battery thermal runaway early warning method and system based on multi-dimensional feature fusion

Abstract

The invention relates to the technical field of battery safety management, in particular to a battery thermal runaway early warning method and system based on multi-dimensional feature fusion. The method comprises the following steps: collecting thermal characteristic data, gas characteristic data and electrochemical characteristic data of a battery; calculating a temperature gradient vector and an isothermal consistency index by using the thermal characteristic data, and capturing abnormal changes of a battery hot spot region by using a local dynamic sampling increasing method; calculating an internal and external field temperature difference phase difference by utilizing the thermal characteristic data and the gas characteristic data, and judging a heat source attribute by combining a heat contribution ratio model; and calculating a cross-modal early response phase difference, constructing a phase difference feature matrix, combining the temperature gradient vector, the heat source attribute and the phase difference feature matrix to form a fusion feature matrix, and inputting the fusion feature matrix into a dynamic weight fusion discrimination model to calculate a thermal runaway risk probability. According to the invention, based on a multi-dimensional feature fusion method, the thermal runaway risk probability is calculated in real time, and the early-stage accurate early warning of the thermal runaway of the battery is realized.

Description

Technical Field

[0001] This invention relates to the field of battery safety management technology, specifically to a battery thermal runaway early warning method and system based on multi-dimensional feature fusion. Background Technology

[0002] With the widespread application of power batteries in new energy vehicles, energy storage systems, and high-power equipment, the number of individual cells and the degree of system integration are constantly increasing, and the threat of thermal runaway accidents to personnel and equipment safety is becoming increasingly prominent. Existing thermal runaway early warning methods mainly rely on single-mode monitoring such as temperature or voltage and fixed threshold judgment. Under complex operating conditions, single parameters are easily affected by changes in the external environment or local anomalies, resulting in insufficient early warning accuracy. Therefore, multi-modal fusion analysis based on thermal characteristics, gas characteristics, and electrochemical characteristics has become a key technical direction for improving the level of battery safety monitoring.

[0003] During battery pack operation, there are still some key issues that are difficult to monitor effectively in certain critical scenarios: First, when the overall temperature distribution on the battery surface tends to be uniform, the temperature rise signal of local early hot spots will be masked, forming isothermal shielding, causing monitoring methods based on temperature uniformity indicators to miss risk points; Second, when external heat sources, such as heat dissipation from adjacent modules, abnormal backflow of the cooling system, or environmental thermal radiation, act on local areas of the battery, they will create abnormal signals in the temperature distribution similar to internal self-heating, forming false hot spots from external heat sources, leading to system misjudgment and triggering unnecessary protection actions; Third, the response time of different physicochemical processes in the early stage of thermal runaway varies, which will cause the time-series correlation characteristics to be ignored in risk assessment. Summary of the Invention

[0004] The purpose of this invention is to provide a battery thermal runaway early warning method and system based on multi-dimensional feature fusion, so as to solve the key problem mentioned in the background art that it is difficult to effectively monitor in some key scenarios.

[0005] To achieve the above objectives, the technical solution of the present invention is: a battery thermal runaway early warning method based on multi-dimensional feature fusion, comprising:

[0006] S1. Based on the synchronous sampling mechanism of internal and external fields, thermal characteristic data, gas characteristic data and electrochemical characteristic data of the battery are collected;

[0007] S2. Calculate the temperature gradient vector and isothermal consistency index using thermal feature data. When the isothermal consistency index is lower than the preset isothermal threshold and there are abnormalities in the non-thermal feature modes, use the local dynamic upsampling method to capture abnormal changes in the battery hot spot area.

[0008] S3. Calculate the phase difference of the internal and external field temperature difference using thermal characteristic data and gas characteristic data, and determine the heat source attribute by combining the heat contribution ratio model, so as to distinguish between external heat sources and internal self-heating of the battery.

[0009] S4. Based on thermal characteristic data, gas characteristic data and electrochemical characteristic data, calculate the phase difference of the early response across modes and construct the phase difference feature matrix. Combine the temperature gradient vector, heat source attributes and phase difference feature matrix to form a fusion feature matrix, and input it into the dynamic weight fusion discrimination model to calculate the probability of thermal runaway risk.

[0010] Preferably, in S1, the internal and external field synchronous sampling mechanism is a multi-modal sensor synchronous triggering acquisition method with a unified global time base and time drift compensation between the inside and outside of the battery pack, which is used to perform consistent synchronous sampling of multi-modal data in the time domain and spatial domain.

[0011] The thermal characteristic data includes battery surface temperature, local temperature gradient, and cooling outlet temperature difference; the gas characteristic data includes combustible gas concentration, release rate, and gas composition ratio; the electrochemical characteristic data includes terminal voltage, operating current, and internal resistance change rate.

[0012] Preferably, in step S2, the isothermal uniformity index is used to represent the uniformity of the temperature field at each monitoring location of the battery, and the specific calculation method is as follows:

[0013] The temperature gradient vector is calculated using a three-dimensional spatial temperature field interpolation algorithm based on thermal characteristic data, and the isothermal consistency index is calculated based on the magnitude distribution of the temperature gradient vector.

[0014] Preferably, in step S2, the preset isothermal threshold is calculated by taking the thermal characteristic data collected by the battery under rated operating conditions, calculating the upper limit of the isothermal consistency index, and then weighting it with the critical value of the isothermal consistency index to obtain the preset isothermal threshold.

[0015] The non-thermal characteristic modes are two types of non-temperature field signal modes: gas characteristic data and electrochemical characteristic data. Gas characteristic data is used to reflect changes in the battery's gas release state, while electrochemical characteristic data is used to reflect changes in the battery's internal electrochemical reactions and conductivity.

[0016] Preferably, in S2, the local dynamic upsampling method is based on the rate of change of the temperature gradient vector and the abnormal mode trigger signal to jointly determine the upsampling region, and temporarily increase the sensor sampling frequency and spatial sampling density within the upsampling region to capture subtle dynamic features of local temperature changes in the early stage of thermal runaway.

[0017] The specific steps for capturing abnormal changes in battery hotspot regions using the local dynamic upsampling method are as follows:

[0018] Calculate the rate of change of the temperature gradient vector and the suspected hotspot area; dynamically adjust the sampling frequency and increase the number of spatial sampling points within the suspected hotspot area; the data collected by the increased spatial sampling points is the upsampled data, calculate the difference within the sliding time window of the upsampled data, extract the short-term temperature rise rate and compare it with the baseline temperature rise rate; when the short-term temperature rise rate is higher than the baseline temperature rise rate for a period of time exceeding the preset duration, it is confirmed that there is an abnormal temperature change in the suspected hotspot area.

[0019] Preferably, the phase difference between the internal and external temperature fields refers to the phase difference between the temperature rise curve of the battery casing surface and the temperature rise curve of the battery cell on the time axis, which represents the time sequence offset relationship between the external temperature rise and the internal temperature rise during the heat conduction process of the battery.

[0020] The calculation of the phase difference between the internal and external field temperatures using thermal and gas characteristic data is as follows:

[0021] Temperature sampling sequences were acquired from the surface of the battery casing and the interior of the battery cells, and time-aligned under the same time reference. The aligned temperature sampling sequences were then denoised and normalized to obtain the external and internal temperature change curves. The maximum correlation lag time between the external and internal temperature change curves was calculated using a cross-correlation function, and the maximum correlation lag time was normalized to the phase difference value. The rate of change of gas characteristic data within the same time window was extracted, and the phase difference value was corrected by combining the time point of gas anomaly occurrence to obtain the final phase difference between the internal and external temperature differences.

[0022] Preferably, in S3, the heat contribution ratio model is a multi-source energy attribution calculation model based on the battery's thermal characteristic data and gas characteristic data, used to calculate the respective heat proportions of external heat sources and battery internal self-heating, and to determine the heat source attributes.

[0023] The heat source attributes determined by the combined heat contribution ratio model are as follows:

[0024] The total heat change of the battery's external and internal fields within the target time window is calculated based on thermal characteristic data; the chemical heat release within the target time window is determined based on gas characteristic data; the chemical heat release is attributed to the battery's internal self-heating component, and the difference in heat change between the external and internal fields is attributed to the external heat source component; the ratio of the internal self-heating component to the total heat change is calculated as the internal heat contribution rate; when the internal heat contribution rate is higher than a set threshold, the heat source attribute is determined to be internal self-heating, otherwise it is determined to be an external heat source.

[0025] Preferably, in S4, the cross-modal early response phase difference refers to the quantitative index of the difference in multimodal response time calculated by the time alignment method and the phase extraction method when the thermal characteristic data, gas characteristic data and electrochemical characteristic data of the battery change, which is used to describe the order of different physicochemical processes in the thermal runaway of the battery.

[0026] The specific method for calculating the phase difference of the early response across modes and constructing the phase difference feature matrix is ​​as follows:

[0027] The thermal, gaseous, and electrochemical characteristic data of the battery are preprocessed and time-synchronized under a unified time reference; the characteristic peak points of each mode signal are extracted; the response time difference between any two modes is calculated and converted into a phase difference value; the phase difference values ​​between any two modes are filled in the mode combination order to construct a phase difference feature matrix.

[0028] Preferably, in step S4, the fused feature matrix is ​​formed by combining the temperature gradient vector, heat source attributes, and phase difference feature matrix in a preset feature arrangement order;

[0029] The dynamic weight fusion discrimination model is constructed using a basic feature weight adaptive allocation algorithm and a nonlinear probability mapping mechanism. It is used to calculate the probability of thermal runaway risk during battery thermal runaway. The specific method is as follows:

[0030] The weight coefficients of each feature in the fusion feature matrix are dynamically calculated. The fusion feature matrix is ​​then distributed according to the weight coefficients to obtain a weighted fusion vector. The weighted fusion vector is then input into a multi-layer nonlinear mapping structure for probability mapping to obtain the probability value of thermal runaway risk.

[0031] On the other hand, the present invention provides a battery thermal runaway early warning system based on multi-dimensional feature fusion, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the battery thermal runaway early warning method based on multi-dimensional feature fusion described above.

[0032] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects:

[0033] 1. In this invention, based on the synchronous acquisition of multimodal sensors and the three-dimensional temperature field gradient analysis, it is possible to identify isothermal shading areas caused by the masking of local temperature gradients when the overall temperature field is relatively uniform, thereby discovering potential abnormal heating points in advance.

[0034] 2. In this invention, by analyzing the heat contribution ratio and calculating the phase difference of the cross-modal response, it is possible to effectively distinguish between internal self-heating of the battery and false hot spots caused by external heat sources. Furthermore, by utilizing the sequential relationship of multiple signals, the accuracy and reliability of thermal runaway risk assessment can be improved. Attached Figure Description

[0035] Figure 1 This is a flowchart of one embodiment of the present invention. Detailed Implementation

[0036] Example 1, as Figure 1 As shown, the present invention proposes a battery thermal runaway early warning method based on multi-dimensional feature fusion, and its specific implementation steps are as follows:

[0037] S1. Based on the synchronous sampling mechanism of internal and external fields, thermal characteristic data, gas characteristic data and electrochemical characteristic data of the battery are collected;

[0038] S2. Calculate the temperature gradient vector and isothermal consistency index using thermal feature data. When the isothermal consistency index is lower than the preset isothermal threshold and there are abnormalities in the non-thermal feature modes, use the local dynamic upsampling method to capture abnormal changes in the battery hot spot area.

[0039] S3. Calculate the phase difference of the internal and external field temperature difference using thermal characteristic data and gas characteristic data, and determine the heat source attribute by combining the heat contribution ratio model, so as to distinguish between external heat sources and internal self-heating of the battery.

[0040] S4. Based on thermal characteristic data, gas characteristic data and electrochemical characteristic data, calculate the phase difference of the early response across modes and construct the phase difference feature matrix. Combine the temperature gradient vector, heat source attributes and phase difference feature matrix to form a fusion feature matrix, and input it into the dynamic weight fusion discrimination model to calculate the probability of thermal runaway risk.

[0041] In this embodiment S1, the internal and external field synchronous sampling mechanism is a multi-modal sensor synchronous triggering acquisition method with unified global time base and time drift compensation between the inside and outside of the battery pack, which is used to perform consistent synchronous sampling of multi-modal data in the time domain and spatial domain.

[0042] The thermal characteristic data includes battery surface temperature, local temperature gradient, and cooling outlet temperature difference; the gas characteristic data includes combustible gas concentration, release rate, and gas composition ratio; the electrochemical characteristic data includes terminal voltage, operating current, and internal resistance change rate.

[0043] In this embodiment, the purpose of the internal and external field synchronous sampling mechanism is to ensure that the external environment monitoring sensors and the internal embedded sensors of the battery pack are strictly aligned in terms of acquisition time, so as to avoid cross-modal data mismatch caused by time deviation. The internal and external field synchronous sampling mechanism achieves unified triggering of all sensors by setting a unified global time base in the system main control unit and combining a temperature-compensated low-drift crystal oscillator and a synchronous pulse trigger signal generation module. For possible sampling clock drift, the system adopts bidirectional time difference measurement and drift compensation algorithm for dynamic correction, thereby ensuring the time synchronization of internal and external field data.

[0044] In terms of sensor deployment, thermal characteristic sensors are preferably distributed in the surface hot spots of the battery module, at the coolant inlet and outlet, to capture changes in temperature gradient and cooling system heat exchange efficiency. Gas sensors are deployed in the gas flow path inside the battery casing and in the near-field area outside the casing to support real-time detection of changes in combustible gas concentration and analysis of released gas components. The electrochemical characteristic acquisition module is connected to the battery management system bus to obtain high-precision terminal voltage, instantaneous operating current, and internal resistance change rate obtained based on multi-frequency AC impedance measurement. After preliminary filtering and quantization processing at the acquisition end, the acquired raw data is appended with a synchronization timestamp and acquisition location information and transmitted to the central processing unit via a low-latency bus.

[0045] In this embodiment S2, the isothermal uniformity index is used to represent the uniformity of the temperature field at each monitoring location of the battery, and the specific calculation method is as follows:

[0046] The temperature gradient vector is calculated using a three-dimensional spatial temperature field interpolation algorithm based on thermal characteristic data, and the isothermal consistency index is calculated based on the magnitude distribution of the temperature gradient vector.

[0047] In this embodiment, the calculation of the temperature gradient vector aims to reflect the spatial distribution characteristics of temperature changes among different monitoring points of the battery pack. Using thermal characteristic data collected synchronously from internal and external fields, a three-dimensional spatial coordinate system of the battery pack is established using a three-dimensional spatial temperature field interpolation algorithm. Based on the physical installation location of each sensor and the measured temperature values, a continuous three-dimensional temperature field is generated using multivariate spline interpolation or radial basis function interpolation methods. In the three-dimensional temperature field, the temperature gradient vector consists of the first-order partial derivatives of the temperature at each coordinate point. The vector direction represents the fastest path of temperature increase, while the vector magnitude characterizes the strength of local temperature changes. Here, the three-dimensional spatial temperature field interpolation algorithm refers to a calculation method that reconstructs the continuous temperature distribution of the battery in three-dimensional space based on discrete temperature sampling points, used to obtain a complete spatial temperature field when the number of sensors is limited.

[0048] In this embodiment, the isothermal consistency index is used to quantify the uniformity of the entire temperature field. Its calculation is based on the statistical distribution of the magnitude of the temperature gradient vector. Specifically, the entire battery pack monitoring area is divided into several spatial units, the mean and variance of the magnitude of the temperature gradient vector in each unit are calculated, and the inverse ratio after variance normalization is evaluated globally as the isothermal consistency index. The isothermal consistency index ranges from 0 to 1. The closer the value is to 1, the more uniform the temperature field distribution is, and there are no significant hot spots inside. Conversely, it indicates that there is a strong local temperature difference phenomenon.

[0049] In this embodiment, to improve anti-interference capability, a temperature sensor calibration compensation and outlier elimination mechanism are introduced in the process of calculating the isothermal consistency index to avoid the amplification effect of single-point measurement error on the isothermal consistency index. The isothermal consistency index is not only used for state judgment at a single moment, but can also be combined with time series analysis to monitor its changing trend, so as to identify in advance the abnormal evolution process of temperature field caused by the decrease of external cooling efficiency or internal local heating.

[0050] In this embodiment S2, the preset isothermal threshold is calculated by taking the thermal characteristic data collected by the battery under rated operating conditions, calculating the upper limit of the isothermal consistency index, and then weighting it with the critical value of the isothermal consistency index to obtain the preset isothermal threshold.

[0051] The non-thermal characteristic modes are two types of non-temperature field signal modes: gas characteristic data and electrochemical characteristic data. Gas characteristic data is used to reflect changes in the battery's gas release state, while electrochemical characteristic data is used to reflect changes in the battery's internal electrochemical reactions and conductivity.

[0052] In this embodiment, the preset isothermal threshold is set based on the thermal characteristic data collected under rated operating conditions. First, stratified sampling is performed according to the ambient temperature range, state of charge range, and cooling condition. A synchronous sampling mechanism for internal and external fields is used to obtain the temperature sequence within a stable period. After sensor zero-point / range calibration and drift correction, outlier removal, and noise reduction are performed on each layer of samples, the three-dimensional temperature field is reconstructed and the population distribution of the isothermal consistency index is calculated. The upper limit of the distribution of each layer is determined by combining the median absolute deviation statistical method. Then, the isothermal consistency critical value from historical thermal runaway events is introduced and weighted by scene weights. A safety margin and hysteresis interval are added to form the preset isothermal threshold. The preset isothermal threshold supports online updates, using a sliding time window and scene recognition to trigger recalculation. A freeze / thaw strategy is enabled when switching operating conditions to avoid frequent jitter. When the sensor fails or the data is incomplete, it reverts to the factory-calibrated threshold and records the event.

[0053] In this embodiment, the non-thermal characteristic modes are divided into two categories: gas characteristic data and electrochemical characteristic data. The gas characteristic data includes combustible gas concentration, release rate, and gas composition ratio. During acquisition, dual-channel constant flow sampling is used, combined with cross-sensitivity compensation and temperature and humidity compensation, and effective segments are screened based on response time and stability threshold. The electrochemical characteristic data includes terminal voltage, operating current, and internal resistance change rate. Preferably, DC internal resistance is acquired under stable load or specified pulse conditions, AC impedance elements are extracted within a limited frequency band, and comparable characteristic segments are generated by combining OCV-SOC mapping and load normalization. Both types of non-thermal characteristic modes, gas characteristic data and electrochemical characteristic data, are aligned with thermal characteristic data under a unified time base.

[0054] In this embodiment S2, the local dynamic upsampling method is based on the rate of change of the temperature gradient vector and the abnormal mode trigger signal to jointly determine the upsampling region, and temporarily increase the sensor sampling frequency and spatial sampling density within the upsampling region to capture subtle dynamic features of local temperature changes in the early stage of thermal runaway.

[0055] The specific steps for capturing abnormal changes in battery hotspot regions using the local dynamic upsampling method are as follows:

[0056] Calculate the rate of change of the temperature gradient vector and the suspected hotspot area; dynamically adjust the sampling frequency and increase the number of spatial sampling points within the suspected hotspot area; the data collected by the increased spatial sampling points is the upsampled data, calculate the difference within the sliding time window of the upsampled data, extract the short-term temperature rise rate and compare it with the baseline temperature rise rate; when the short-term temperature rise rate is higher than the baseline temperature rise rate for a period of time exceeding the preset duration, it is confirmed that there is an abnormal temperature change in the suspected hotspot area.

[0057] In this embodiment, the method for calculating the rate of change of the temperature gradient vector, in the context of battery thermal runaway early warning, refers to the mathematical process of quantifying the rate of change of the temperature field over a time series, used to identify whether there are signs of accelerated temperature rise in hot spots. The logic for calculating the rate of change of the temperature gradient vector is to first calculate the temperature gradient vector at each time segment, and then compare the gradient change magnitude between adjacent time segments. Specifically, at each sampling moment, the temperature gradient vector is calculated based on the interpolation results of the three-dimensional spatial temperature field; the difference in magnitude of the gradient vectors at adjacent moments is divided by the sampling period to obtain the rate of change of the temperature gradient vector.

[0058] In this embodiment, the local dynamic upsampling method is used to temporarily collect high-density data in a specific area when an abnormal trend in thermal feature data is detected, so as to improve the accuracy of capturing early signs of thermal runaway. The local dynamic upsampling method first relies on the thermal feature data stream under a unified time base to calculate the rate of change of the temperature gradient vector in a continuous sampling period. The spatial range of suspected hotspot areas is identified by the peak distribution of the rate of change and the local clustering method. At the same time, the trigger signal of non-thermal feature mode is combined to eliminate false hotspots caused by environmental disturbances or measurement noise, so as to ensure the effectiveness of the upsampling area location. After the upsampling area is determined, the temperature sensor or sensor array in the relevant area is controlled to temporarily increase the sampling frequency, for example, the sampling frequency is increased to twice the original sampling frequency, and additional measurement points are introduced in space. These additional measurement points can be implemented by enabling multi-channel measurement by nearby sensors or by simulation through virtual measurement point interpolation algorithm, thereby increasing the spatial sampling density.

[0059] The collected upsampled data is subjected to sliding difference operation within a unified time window to extract the short-term temperature rise rate curve. This curve is then compared point by point with a pre-calibrated baseline temperature rise rate. A duration determination strategy is used to distinguish between occasional disturbances and continuous abnormal temperature rise. If it is determined to be a continuous abnormal temperature rise, the area is marked as a high-risk hotspot area. The duration determination strategy is a time threshold strategy for judging when the short-term temperature rise rate is continuously higher than the baseline temperature rise rate.

[0060] In this embodiment S3, the phase difference between the internal and external temperature fields refers to the phase difference between the temperature rise curve of the battery casing surface and the temperature rise curve of the battery cell on the time axis. It represents the time sequence offset relationship between the external temperature rise and the internal temperature rise during the heat conduction process of the battery.

[0061] The calculation of the phase difference between the internal and external field temperatures using thermal and gas characteristic data is as follows:

[0062] Temperature sampling sequences were acquired from the surface of the battery casing and the interior of the battery cells, and time-aligned under the same time reference. The aligned temperature sampling sequences were then denoised and normalized to obtain the external and internal temperature change curves. The maximum correlation lag time between the external and internal temperature change curves was calculated using a cross-correlation function, and the maximum correlation lag time was normalized to the phase difference value. The rate of change of gas characteristic data within the same time window was extracted, and the phase difference value was corrected by combining the time point of gas anomaly occurrence to obtain the final phase difference between the internal and external temperature differences.

[0063] In this embodiment, the phase difference of the internal and external temperature differences is used to quantitatively characterize the time delay between internal heating and external temperature rise during the heat conduction process of the battery, thereby providing a basis for judging the heat source attributes. The physical meaning of the phase difference of the internal and external temperature differences is that when the heat mainly comes from the external heat source, the temperature rise on the outer shell surface usually occurs before the temperature rise inside the battery; while when the heat mainly comes from the self-heating inside the battery, the internal temperature rise will occur before the external temperature rise. The sign and absolute value of the phase difference can directly reflect the difference in this heat conduction path.

[0064] In this embodiment, the cross-correlation function is a mathematical tool for measuring the similarity of two signals under different time delays. It is used to find the optimal alignment position of the two signals on the time axis, i.e., the maximum correlation lag time, thereby quantifying their temporal relationship. In the battery thermal runaway early warning scenario, the cross-correlation function can be used to compare the external temperature rise curve and the internal temperature rise curve to find the time offset of their temperature rise changes, which is the phase difference between the internal and external temperature differences.

[0065] In this embodiment, the calculation process for the phase difference between the internal and external temperature fields includes the following detailed steps: Based on a unified global time base, external temperature sampling sequences and internal temperature sampling sequences are obtained from temperature sensors deployed on the surface of the battery casing and inside the battery cells, respectively, and time synchronization and sampling frequency alignment are performed; wavelet threshold denoising is used to denoise the sampling sequences to suppress measurement noise and instantaneous disturbances, and then the absolute temperature deviation between different measurement points is eliminated by interval normalization to obtain the external temperature change curve and the internal temperature change curve; in the signal processing stage, the correlation coefficient of the two temperature change curves under different time delays is analyzed using the cross-correlation function, the lag time corresponding to the maximum value of the correlation coefficient is determined, and the lag time is normalized with the sampling period to obtain the phase difference value.

[0066] In this embodiment, to avoid misjudgment of phase difference caused by short-term gas leakage or abnormal gas release, gas characteristic data is introduced as a correction factor: the rate of change of gas concentration is extracted within the same time window, and the time point when the gas anomaly occurs is located. This time point is used as a weight correction parameter to adjust the original phase difference value, thereby obtaining the final internal and external field temperature phase difference that can reflect the true temporal relationship of the heat source.

[0067] In this embodiment S3, the heat contribution ratio model is a multi-source energy attribution calculation model based on the battery's thermal characteristic data and gas characteristic data. It is used to calculate the respective heat ratios of external heat sources and battery internal self-heating, and to determine the heat source attributes.

[0068] The heat source attributes determined by the combined heat contribution ratio model are as follows:

[0069] The total heat change of the battery's external and internal fields within the target time window is calculated based on thermal characteristic data; the chemical heat release within the target time window is determined based on gas characteristic data; the chemical heat release is attributed to the battery's internal self-heating component, and the difference in heat change between the external and internal fields is attributed to the external heat source component; the ratio of the internal self-heating component to the total heat change is calculated as the internal heat contribution rate; when the internal heat contribution rate is higher than a set threshold, the heat source attribute is determined to be internal self-heating, otherwise it is determined to be an external heat source.

[0070] In this embodiment, the heat contribution ratio model is used to distinguish the relative contribution ratios of internal self-heating and external heat source input to the overall heat change in the early stage of battery thermal runaway, thereby providing a targeted response basis for early warning strategies. The heat contribution ratio model uses thermal characteristic data and gas characteristic data to establish a multi-source energy attribution calculation framework, and realizes the quantitative determination of heat source attributes through the constraints of energy conservation relationship and chemical reaction heat release. In the specific implementation process, firstly, under a unified time reference, a target time window is selected, and the temperature data of the battery shell surface and the battery cell internal are integrated and converted to obtain the total external heat change and the total internal heat change, respectively. Based on the gas characteristic data and the known chemical reaction heat parameters of the battery, the chemical heat release within the target time window is calculated and directly attributed to the internal self-heating component.

[0071] In this embodiment, the target time window refers to a preset time range extended forward or backward from the initial time point of the detected thermal runaway risk of the battery. This time range is used to statistically analyze the battery's thermal and gas characteristic data for thermal attribution calculations. The starting point of the target time window is determined by a triggering condition, such as an isothermal consistency index below a threshold, a temperature gradient vector change rate exceeding a set value, or abnormal gas concentration. The length of the target time window can be fixed or adaptive. The target time window must maintain a consistent time base across multimodal data. The target time window is a preset duration interval extended forward or backward from the starting time point of the triggering event when a thermal runaway risk is detected, or an adaptively adjusted time interval based on the data change rate, used to simultaneously statistically analyze thermal and gas characteristic data within this time period.

[0072] In this embodiment, to distinguish the effect of external heat sources, the difference between the total heat change in the external field and the internal field is regarded as the heat component introduced by the external heat source. At the same time, the external heat source component is compensated and corrected for changes in ambient temperature and heat dissipation conditions to avoid overestimation or underestimation due to environmental fluctuations. The ratio of the internal self-heating component to the total heat change is calculated to obtain the internal heat contribution rate. When the contribution rate is higher than a preset ratio threshold, the heat source attribute is determined to be internal self-heating; otherwise, it is determined to be an external heat source. The preset ratio threshold is determined based on experimental calibration or historical data statistics.

[0073] In this embodiment S4, the cross-modal early response phase difference refers to the quantitative index of the difference in multimodal response time calculated by the time alignment method and the phase extraction method when the thermal characteristic data, gas characteristic data and electrochemical characteristic data of the battery change. It is used to describe the order of different physicochemical processes in the thermal runaway of the battery.

[0074] The specific method for calculating the phase difference of the early response across modes and constructing the phase difference feature matrix is ​​as follows:

[0075] The thermal, gaseous, and electrochemical characteristic data of the battery are preprocessed and time-synchronized under a unified time reference; the characteristic peak points of each mode signal are extracted; the response time difference between any two modes is calculated and converted into a phase difference value; the phase difference values ​​between any two modes are filled in the mode combination order to construct a phase difference feature matrix.

[0076] In this embodiment, the introduction of cross-modal early response phase difference aims to reveal the sequential relationship and interaction between thermal, gaseous, and electrochemical characteristics during battery thermal runaway through time-domain quantification. In the specific calculation of the cross-modal early response phase difference, the three types of characteristic data are first preprocessed under a unified time reference frame, including denoising, amplitude normalization, and baseline drift correction. A dynamic time warping timing synchronization method is used to ensure the comparability of key change nodes for each mode at the same reference time. In the peak feature point extraction stage… Characteristic peak identification criteria are set for the physical characteristics of different modal signals. For example, thermal characteristic data can be identified by the moment when the first temperature rise rate reaches a set threshold, gas characteristic data can be identified by the moment when the change rate of combustible gas concentration reaches its peak, and electrochemical characteristic data can be identified by the moment when the change rate of terminal voltage or internal resistance deviates significantly from the steady state. After obtaining the peak points of each mode, the time difference between any two mode peaks is calculated and converted into a phase difference value. These phase difference values ​​are filled into a phase difference feature matrix in the order of mode combination, such as thermal-gas, thermal-electrochemical, and gas-electrochemical.

[0077] In this embodiment S4, the fused feature matrix is ​​formed by combining the temperature gradient vector, heat source attributes, and phase difference feature matrix in a preset feature arrangement order;

[0078] The dynamic weight fusion discrimination model is constructed using a basic feature weight adaptive allocation algorithm and a nonlinear probability mapping mechanism. It is used to calculate the probability of thermal runaway risk during battery thermal runaway. The specific method is as follows:

[0079] The weight coefficients of each feature in the fusion feature matrix are dynamically calculated. The fusion feature matrix is ​​then distributed according to the weight coefficients to obtain a weighted fusion vector. The weighted fusion vector is then input into a multi-layer nonlinear mapping structure for probability mapping to obtain the probability value of thermal runaway risk.

[0080] In this embodiment, the construction of the fusion feature matrix aims to integrate three core features—temperature gradient vector, heat source attributes, and phase difference feature matrix—in an orderly manner to form a multi-dimensional input set covering spatial distribution features, energy source features, and cross-modal temporal features. The core of the dynamic weight fusion discrimination model lies in realizing real-time adaptive adjustment of feature weights, rather than using fixed weights. This can cope with changes in feature importance caused by differences in battery operating status, environmental conditions, and sensor distribution. The feature weight adaptive allocation algorithm can evaluate the sensitivity of each feature to the thermal runaway discrimination result based on the joint statistical results of historical samples and real-time data, using gradient updates, entropy weighting, or attention mechanisms, and dynamically allocate weight coefficients accordingly.

[0081] In this embodiment, in the nonlinear probability mapping mechanism, a multi-layer neural network structure is used to map the weighted fusion vector to the risk probability value in the [0,1] interval. The nonlinear probability mapping mechanism not only considers the linear superposition relationship between features, but also captures the higher-order interaction between features through a nonlinear activation function. For example, the joint anomaly of temperature gradient change and phase difference feature is often a precursor to thermal runaway. The nonlinear mapping mechanism can amplify the risk response under such multi-feature combination.

[0082] Example 2: The present invention proposes a battery thermal runaway early warning system based on multi-dimensional feature fusion, which is applied to the battery thermal runaway early warning method based on multi-dimensional feature fusion proposed in Example 1. It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the battery thermal runaway early warning method based on multi-dimensional feature fusion in Example 1.

[0083] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A battery thermal runaway early warning method based on multi-dimensional feature fusion, characterized in that, Includes the following steps: S1. Based on the synchronous sampling mechanism of internal and external fields, thermal characteristic data, gas characteristic data and electrochemical characteristic data of the battery are collected; S2. Calculate the temperature gradient vector and isothermal consistency index using thermal feature data. When the isothermal consistency index is lower than the preset isothermal threshold and there are abnormalities in the non-thermal feature modes, use the local dynamic upsampling method to capture abnormal changes in the battery hot spot area. S3. Calculate the phase difference of the internal and external field temperature difference using thermal characteristic data and gas characteristic data, and determine the heat source attribute by combining the heat contribution ratio model, so as to distinguish between external heat sources and internal self-heating of the battery. S4. Based on thermal characteristic data, gas characteristic data and electrochemical characteristic data, calculate the phase difference of the early response across modes and construct the phase difference feature matrix. Combine the temperature gradient vector, heat source attributes and phase difference feature matrix to form a fusion feature matrix, and input it into the dynamic weight fusion discrimination model to calculate the probability of thermal runaway risk.

2. The battery thermal runaway early warning method based on multi-dimensional feature fusion according to claim 1, characterized in that: In S1, the internal and external field synchronous sampling mechanism is a multi-modal sensor synchronous triggering acquisition method with unified global time base and time drift compensation between the inside and outside of the battery pack, which is used to perform consistent synchronous sampling of multi-modal data in the time domain and spatial domain. The thermal characteristic data includes battery surface temperature, local temperature gradient, and cooling outlet temperature difference; the gas characteristic data includes combustible gas concentration, release rate, and gas composition ratio; the electrochemical characteristic data includes terminal voltage, operating current, and internal resistance change rate.

3. The battery thermal runaway early warning method based on multi-dimensional feature fusion according to claim 2, characterized in that: In step S2, the isothermal uniformity index is used to represent the uniformity of the temperature field at each monitoring location of the battery. The specific calculation method is as follows: The temperature gradient vector is calculated using a three-dimensional spatial temperature field interpolation algorithm based on thermal characteristic data, and the isothermal consistency index is calculated based on the magnitude distribution of the temperature gradient vector.

4. The battery thermal runaway early warning method based on multi-dimensional feature fusion according to claim 3, characterized in that: In S2, the preset isothermal threshold is calculated by taking the thermal characteristic data collected by the battery under rated operating conditions, calculating the upper limit of the isothermal consistency index, and then weighting it with the critical value of the isothermal consistency index to obtain the preset isothermal threshold. The non-thermal characteristic modes are two types of non-temperature field signal modes: gas characteristic data and electrochemical characteristic data. Gas characteristic data is used to reflect changes in the battery's gas release state, while electrochemical characteristic data is used to reflect changes in the battery's internal electrochemical reactions and conductivity.

5. The battery thermal runaway early warning method based on multi-dimensional feature fusion according to claim 4, characterized in that: In S2, the local dynamic upsampling method is based on the rate of change of the temperature gradient vector and the abnormal mode trigger signal to jointly determine the upsampling region, and temporarily increase the sensor sampling frequency and spatial sampling density in the upsampling region to capture the subtle dynamic features of local temperature changes in the early stage of thermal runaway. The specific steps for capturing abnormal changes in battery hotspot regions using the local dynamic upsampling method are as follows: Calculate the rate of change of the temperature gradient vector and the suspected hotspot area; dynamically adjust the sampling frequency and increase the number of spatial sampling points within the suspected hotspot area; the data collected by the increased spatial sampling points is the upsampled data, calculate the difference within the sliding time window of the upsampled data, extract the short-term temperature rise rate and compare it with the baseline temperature rise rate; when the short-term temperature rise rate is higher than the baseline temperature rise rate for a period of time exceeding the preset duration, it is confirmed that there is an abnormal temperature change in the suspected hotspot area.

6. The battery thermal runaway early warning method based on multi-dimensional feature fusion according to claim 5, characterized in that: The phase difference between the internal and external temperature fields refers to the phase difference between the temperature rise curve of the battery casing surface and the temperature rise curve of the battery cell interior on the time axis. It represents the time sequence offset relationship between the external temperature rise and the internal temperature rise during the battery heat conduction process. The calculation of the phase difference between the internal and external field temperatures using thermal and gas characteristic data is as follows: Temperature sampling sequences were acquired from the surface of the battery casing and the interior of the battery cells, and time-aligned under the same time reference. The aligned temperature sampling sequences were then denoised and normalized to obtain the external and internal temperature change curves. The maximum correlation lag time between the external and internal temperature change curves was calculated using a cross-correlation function, and the maximum correlation lag time was normalized to the phase difference value. The rate of change of gas characteristic data within the same time window was extracted, and the phase difference value was corrected by combining the time point of gas anomaly occurrence to obtain the final phase difference between the internal and external temperature differences.

7. The battery thermal runaway early warning method based on multi-dimensional feature fusion according to claim 6, characterized in that: In S3, the heat contribution ratio model is a multi-source energy attribution calculation model based on the battery's thermal characteristic data and gas characteristic data. It is used to calculate the respective heat ratios of external heat sources and battery internal self-heating, and to determine the heat source attributes. The heat source attributes determined by the combined heat contribution ratio model are as follows: The total heat change of the battery's external and internal fields within the target time window is calculated based on thermal characteristic data; the chemical heat release within the target time window is determined based on gas characteristic data; the chemical heat release is attributed to the battery's internal self-heating component, and the difference in heat change between the external and internal fields is attributed to the external heat source component; the ratio of the internal self-heating component to the total heat change is calculated as the internal heat contribution rate; when the internal heat contribution rate is higher than a set threshold, the heat source attribute is determined to be internal self-heating, otherwise it is determined to be an external heat source.

8. The battery thermal runaway early warning method based on multi-dimensional feature fusion according to claim 7, characterized in that: In S4, the cross-modal early response phase difference refers to the quantitative index of the difference in multimodal response time calculated by the time alignment method and the phase extraction method when the thermal characteristic data, gas characteristic data and electrochemical characteristic data of the battery change. It is used to describe the order of different physicochemical processes in the thermal runaway of the battery. The specific method for calculating the phase difference of the early response across modes and constructing the phase difference feature matrix is ​​as follows: The thermal, gaseous, and electrochemical characteristic data of the battery are preprocessed and time-synchronized under a unified time reference; the characteristic peak points of each mode signal are extracted; the response time difference between any two modes is calculated and converted into a phase difference value; the phase difference values ​​between any two modes are filled in the mode combination order to construct a phase difference feature matrix.

9. A battery thermal runaway early warning method based on multi-dimensional feature fusion according to claim 8, characterized in that: In step S4, the fusion feature matrix is ​​formed by combining the temperature gradient vector, heat source attributes, and phase difference feature matrix according to a preset feature arrangement order. The dynamic weight fusion discrimination model is constructed using a basic feature weight adaptive allocation algorithm and a nonlinear probability mapping mechanism. It is used to calculate the probability of thermal runaway risk during battery thermal runaway. The specific method is as follows: The weight coefficients of each feature in the fusion feature matrix are dynamically calculated. The fusion feature matrix is ​​then distributed according to the weight coefficients to obtain a weighted fusion vector. The weighted fusion vector is then input into a multi-layer nonlinear mapping structure for probability mapping to obtain the probability value of thermal runaway risk.

10. A battery thermal runaway early warning system based on multi-dimensional feature fusion, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the battery thermal runaway early warning method based on multi-dimensional feature fusion as described in any one of claims 1-9.

Images