A lithium ion battery thermal runaway active inhibition method and system based on electrical regulation

Abstract

The application belongs to the field of battery thermal runaway inhibition, and provides a lithium ion battery thermal runaway active inhibition method and system based on electrical regulation. Based on the index correlation empowerment method, weights are applied to the candidate indexes, the thermal runaway severity comprehensive score of each index combination is calculated, each index combination is quantitatively evaluated from three dimensions of sampling noise robustness, sample disturbance robustness and severity distinguishing ability, a multi-objective optimization method is used to select the optimal index combination; taking the working condition variables including thermal abuse thermal power, battery SOC and discharge rate as input, using the pre-trained thermal runaway severity prediction model based on Gaussian process regression, the corresponding prediction mean and prediction standard deviation are obtained; according to the upper confidence limit, the active inhibition safety boundary is deduced, the working condition variables corresponding to the thermal runaway severity comprehensive score within the safety boundary are searched, and the active inhibition strategy action target is determined. The application realizes the delay or even blocking of thermal runaway.

Description

Technical Field

[0001] This invention belongs to the field of battery thermal runaway suppression, specifically relating to an active suppression method and system for thermal runaway of lithium-ion batteries based on electro-regulation. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Lithium-ion batteries, characterized by high energy density, long cycle life, and environmental friendliness, are widely used in energy storage systems and electric vehicles. However, frequent battery thermal runaway incidents have caused significant economic losses. In reality, battery systems operate under complex and variable conditions, potentially exceeding the limits of their existing safety features. This can lead to batteries being abused without being detected, leaving them on the verge of danger. As individual battery cell failures continue to evolve, heat accumulates through positive feedback, eventually resulting in sudden thermal runaway. This exposes neighboring cells to extreme thermal abuse, easily triggering a chain reaction of thermal runaway. At this point, emergency measures become the last line of defense, aiming to prevent the spread of heat and avoid escalating individual cell incidents into catastrophic fires and explosions at the battery pack level.

[0004] Current emergency measures primarily rely on pressure relief, heat insulation, and fire extinguishing, or a combination of all three. However, whether it's pressure relief at the individual or module level, or filling the spaces between cells with insulating materials, these are all passive protection methods lacking active control capabilities. They can only allow the situation to develop unchecked and are unlikely to effectively reverse an out-of-control situation. Fire extinguishing and cooling methods, such as spraying fire extinguishing agents, ignore the inherent nature of the continuous heat generation from the internal chemical reactions of thermal runaway batteries. They only address the symptoms, not the root cause, and cannot suppress heat accumulation at its source. Furthermore, they are difficult to deploy and implement in space-constrained environments. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes an active suppression method and system for thermal runaway in lithium-ion batteries based on electro-regulation. This invention achieves efficient delaying or even blocking of thermal runaway.

[0006] According to some embodiments, the present invention adopts the following technical solution: A method for actively suppressing thermal runaway in lithium-ion batteries based on electro-regulation includes the following steps: Based on the index correlation weighting method, weights are applied to the physical field parameters that reflect the severity of battery thermal runaway, and the comprehensive score of thermal runaway severity of each index combination is calculated. The index combination is quantitatively evaluated from three dimensions: sampling noise robustness, sample perturbation robustness, and severity discrimination ability. A multi-objective optimization method is used to select the index combination with the best comprehensive score of thermal runaway severity. Using operating condition variables including thermal abuse power, battery SOC and discharge rate as inputs, a pre-trained thermal runaway severity prediction model based on Gaussian process regression is used to obtain the mean and standard deviation of the comprehensive score prediction for the thermal runaway severity of the optimal combination of indicators. Based on the predicted mean and predicted standard deviation, the active suppression safety boundary is inferred from the upper confidence bound. The operating condition variable corresponding to the comprehensive score of thermal runaway severity within the safety boundary is searched and determined as the action target of the active suppression strategy.

[0007] As an alternative implementation method, the process of assigning weights to physical field parameters reflecting the severity of battery thermal runaway based on the index correlation weighting method and calculating the comprehensive score of thermal runaway severity for each index combination includes: assigning weights to each index using the index correlation weighting method; for candidate indices in each test sample, calculating the normalized value and standard deviation of the candidate indices based on their original values ​​and the maximum and minimum values ​​of each index in all test samples; then calculating the correlation coefficient between the two indices; calculating the information content of each index based on the correlation coefficient; then calculating the objective weight; and finally calculating the comprehensive score vector of thermal runaway severity based on the objective weight. S : ; in, S i For the first i A comprehensive score reflecting the severity of thermal runaway for each test sample. w j For objective weighting, n The total number of test samples, m The total number of candidate indicators in each sample. The normalized value of the candidate indicator; for m There are 2 candidate indicators, which can generate a total of 2 m -1 candidate indicator combination, and for all candidate indicator combinations, calculate the comprehensive score of thermal runaway severity.

[0008] As an alternative implementation, the process of quantitatively evaluating the combination of indicators from the perspective of sampling noise robustness includes: calculating the weight of each indicator using the raw data, and then, for the normalized values ​​of all indicators, repeatedly injecting noise intensity as... σGaussian white noise is used. After each noise injection, a comprehensive score array of thermal runaway severity is calculated based on the noise data, and finally a comprehensive score matrix of thermal runaway severity is obtained. For each column, the coefficient of variation of each test sample under noise perturbation is calculated. The mean of all coefficients of variation is calculated as the sampling noise robustness evaluation index. The smaller the mean, the less sensitive the comprehensive score of thermal runaway severity obtained by the corresponding index combination is to index perturbation, and the stronger the noise robustness.

[0009] As an alternative implementation method, the process of quantitatively evaluating the combination of indicators from the perspective of sample perturbation robustness includes: performing Bootstrap resampling with replacement on the original experimental samples, with the same sample size. n The sample set is repeated multiple times to form multiple Bootstrap sample sets, each containing... n For each Bootstrap sample set, the weights of each indicator are recalculated and reapplied to the original experimental samples to calculate the comprehensive score array of thermal runaway severity. One row represents the comprehensive score array of thermal runaway severity for all experimental samples under a certain Bootstrap resampling weighting, and one column represents all experimental samples under Bootstrap resampling weighting. i The severity of thermal runaway is comprehensively scored. For each column, the coefficient of variation of each test sample under Bootstrap resampling weighting is calculated. The mean of all coefficients of variation is calculated as the sample perturbation robustness evaluation index. The smaller the value, the less sensitive the corresponding index combination is to sample sampling perturbation and the stronger the robustness under random sample set.

[0010] As an alternative implementation method, the process of quantitatively evaluating each combination of indicators from three dimensions of severity discrimination ability includes: using Gini average distance to quantify the discrimination between different sample scores of the comprehensive score of thermal runaway severity calculated by different combinations of indicators. The larger the value, the greater the distance between the values ​​in the SRS array, the greater the discrimination, and the more beneficial it is for subsequent modeling calculations.

[0011] As an alternative implementation method, the process of selecting the optimal combination of indicators for the severity of thermal runaway using a multi-objective optimization method includes: normalizing the quantitative evaluation results of all candidate indicator combinations from three dimensions—sampling noise robustness, sample perturbation robustness, and severity discrimination ability—and then unifying them according to the principle that larger values ​​are better, thus obtaining a multi-objective optimized evaluation indicator array. a , b Two sets of optimization evaluation indicators, if a Group relative b Group satisfies a All quantitative evaluation results of the group were greater than or equal to bA group is defined as a group whose quantitative evaluation results are strictly greater than a certain threshold, and at least one of these results is greater than a certain threshold. a Dominant Group b All groups not dominated by other groups constitute the Pareto front. In the Pareto front, the final candidate index combination is selected by using the mass loss constraint and the principle of the closest ideal point.

[0012] As an alternative implementation, the thermal runaway severity prediction model based on Gaussian process regression takes thermal abuse power, battery SOC and discharge rate as inputs and the observed value of the comprehensive score of thermal runaway severity as output, and uses Gaussian regression process to represent the mapping relationship between input and output.

[0013] As an alternative implementation, the process of inferring the active suppression safety boundary based on the predicted mean and predicted standard deviation, according to the upper confidence bound, includes: based on the predicted mean... u s,i and the predicted standard deviation σ s,i By introducing a one-sided confidence constraint, a predetermined percentage upper confidence bound for the comprehensive score of thermal runaway severity is calculated. S U = u s,i + α ·σ s,i , α Let quantile be a cumulative probability quantile in a standard normal distribution, when the predicted mean is... S U Less than the preset threshold S th At that time, it was considered that thermal runaway could be avoided, and it was regarded as a safe zone, and then the safe boundary surface was determined.

[0014] As an alternative implementation method, the process of determining the operating condition variables corresponding to the comprehensive score of the severity of thermal runaway within the safety boundary as the action target of the active suppression strategy includes: after constructing the safety boundary, for any operating condition of thermal abuse power, battery SOC and discharge rate, firstly determine whether it is in the safe zone according to the safety boundary, and then decide whether it is necessary to intervene. If it is outside the safe zone, for any of the operating condition variables of thermal abuse power, battery SOC and discharge rate that are controllable, search the active suppression space based on the safety boundary and determine the active suppression strategy.

[0015] An active suppression system for thermal runaway in lithium-ion batteries based on electro-regulation, comprising: The optimal index combination determination module is configured to apply weights to physical field parameters that reflect the severity of battery thermal runaway based on the index correlation weighting method, calculate the comprehensive score of thermal runaway severity of each index combination, quantitatively evaluate each index combination from three dimensions: sampling noise robustness, sample perturbation robustness, and severity discrimination ability, and select the index combination with the optimal comprehensive score of thermal runaway severity using a multi-objective optimization method. The thermal runaway severity prediction module is configured to take operating condition variables including thermal abuse power, battery SOC and discharge rate as input, and use a pre-trained thermal runaway severity prediction model based on Gaussian process regression to obtain the mean and standard deviation of the comprehensive score prediction of thermal runaway severity with the optimal combination of indicators. The active suppression strategy search module is configured to, based on the predicted mean and predicted standard deviation, back-calculate the active suppression safety boundary according to the upper confidence bound, search for the operating condition variable corresponding to the comprehensive score of thermal runaway severity within the safety boundary, and determine it as the active suppression strategy action target.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention is based on the fact that thermal runaway is the process of chemical energy being converted into thermal energy and released rapidly. It actively regulates the energy release source inside the battery to achieve the transformation of thermal runaway suppression from "passive protection" to "active suppression" and from "surface weakening" to "source suppression".

[0017] This invention fully considers the multi-field coupling characteristics of thermal runaway, and uses objective weighting and multi-objective optimization methods to integrate multi-field characteristics, providing a scientific and reliable quantitative benchmark for the severity of thermal runaway.

[0018] This invention establishes a thermal runaway severity prediction model based on Gaussian process regression, which can then determine a more conservative and reliable safety boundary based on the upper confidence bound, effectively reducing the safety misjudgment rate. Furthermore, an active suppression strategy is formulated based on an adjustable safety threshold, significantly improving the flexibility of regulation.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a schematic diagram of a method for actively suppressing thermal runaway in a lithium-ion battery according to one embodiment. Figure 2The three-dimensional evaluation results are as follows: (a) is robustness to sampling noise, (b) is robustness to sample perturbation, and (c) is the ability to distinguish severity. Figure 3 This is a schematic diagram of the Pareto front combination in one embodiment; Figure 4 A schematic diagram of an optimal SRS sequence for one embodiment; Figure 5(a) shows the prediction results of the GPR model under the leave-one-out method in one embodiment; Figure 5(b) is a schematic diagram of the residual analysis results of one embodiment; Figure 5(c) is a schematic diagram of the accuracy index results of one embodiment; Figure 6(a) is a schematic diagram showing that all operating condition variables are controllable in one embodiment; Figure 6(b) shows the partially controllable operating condition variables in one embodiment. SOC (0 is fixed at 70%) Schematic diagram. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0025] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0026] Example 1 Active suppression methods for thermal runaway in lithium-ion batteries based on electro-regulation, such as Figure 1As shown, a controllable discharge current is applied to the battery during thermal abuse to actively suppress thermal runaway. First, a comprehensive evaluation index for the severity of thermal runaway, characterized by multi-field coupling, is constructed based on the index correlation weighting method and multi-objective optimization. Then, using operating condition variables such as thermal abuse power, battery SOC, and discharge rate as inputs, a prediction model for the severity of thermal runaway based on Gaussian process regression is established. Finally, the active suppression safety boundary is inferred from the upper confidence bound, and an active suppression strategy is formulated within the controllable space to delay or even block thermal runaway.

[0027] First, this embodiment defines the thermal abuse conditions of the battery, including thermal power, state of charge (SOC), and discharge rate. After being applied to the battery, the battery's temperature, voltage, impact force, and mass loss after failure are collected by sensors.

[0028] Thermal runaway is essentially a multi-field coupled process involving electricity, heat, force, and mass release. Relying on a single parameter may be insufficient to fully capture the complete information regarding the severity of thermal runaway. Therefore, it is necessary to integrate multiple parameters to comprehensively evaluate the severity of thermal runaway.

[0029] Assume there exist candidate indicators [Φ] j , j ∈ These indicators are multi-physics parameters measured by sensors during thermal runaway, reflecting the severity of thermal runaway from different perspectives. However, there may be information redundancy and complementarity among the indicators, thus requiring further fusion processing. A weighting method based on indicator correlation is used to assign weights to each indicator. n There are 10 test samples, and each sample contains 100 samples. m The candidate indicator is denoted as . i In the nth sample j The original values ​​of each indicator are x ij First, calculate its normalized value. : (1) in, x j For the first j Each indicator corresponds to an array of all test samples, with a length of [value missing]. n In the formula, the denominator represents the first... j The difference between the maximum and minimum values ​​of each index across all test samples. Then, calculate the... j Standard deviation of each indicator δ j : (2) in For the first jThe mean of the normalized values ​​of each indicator. Further calculation of the indicators. j With indicators k Correlation coefficient between: (3) Then the first j Information content of each indicator C j Defined as: (4) Furthermore, the first j Objective weight of each indicator w j Represented as: (5) Finally, calculate all according to the following formula. n A number of test samples based on m A vector of combined scores for thermal runaway severity (STRS) for each candidate metric. S : (6) in, S i For the first i STRS of one test sample.

[0030] It should be noted that, for m There are 2 candidate indicators, which can generate a total of 2 m -1 candidate index combinations. For all combinations, the SRS is calculated. Then, each combination is quantitatively evaluated from three dimensions: sampling noise robustness, sample perturbation robustness, and severity discrimination ability, in order to select the optimal combination.

[0031] 1) Sampling noise robustness First, calculate the weights of each indicator using the raw data. Then, normalize the values ​​for all indicators. The intensity of repeated injection noise is σ Gaussian white noise l Each time noise is injected, the SRS array is calculated based on the noise data, ultimately yielding the SRS matrix. S noise : (7) visible, S noise yes l OK n A column matrix, where one row represents the SRS array of all test samples under a given noise injection, and one column represents the test samples under all noise inputs.i The STRS. For each column, calculate the coefficient of variation for each test sample under noise perturbation. : (8) in, ε For small positive numbers, avoid having a denominator of zero in the formula. Finally, calculate all... mean As an evaluation index for sampling noise robustness. The smaller the value, the less sensitive the STRS obtained by the candidate index combination is to index perturbations, and the stronger its noise robustness.

[0032] 2) Sample perturbation robustness Bootstrap resampling with replacement was performed on the original experimental samples, with the same sample size. n The sample set. Repeated l The next step constitutes l Group Bootstrap sample sets, each group contains n Each Bootstrap sample set is tested. For each sample set, the weights of each indicator are recalculated and applied back to the original sample set to calculate the SRS array. S .

[0033] This yields a SRS matrix identical in form to equation (7), where one row represents the SRS array of all test samples under a certain Bootstrap resampling weighting, and one column represents all test samples under Bootstrap resampling weighting. i The SRS. Similarly, for each column, the coefficient of variation of each experimental sample under Bootstrap resampling weighting is calculated according to Equation (8). , and then calculate all mean As an indicator of sample perturbation robustness, the smaller the value, the less sensitive the indicator combination is to sample sampling perturbations, and the stronger its robustness under random sample sets.

[0034] 3) Ability to differentiate severity The Gini mean distance was used to quantify the discriminative power between different sample scores of the STRS calculated from different combinations of indicators. (9) As can be seen, this index calculates the average of the absolute differences between any two SRS values ​​in the SRS array. The larger the value, the greater the distance between the values ​​in the SRS array, the greater the discriminative power, and the more beneficial it is for subsequent modeling calculations.

[0035] It is foreseeable that different combinations of candidate indicators may present contradictions and trade-offs in the three-dimensional evaluation results. Therefore, a multi-objective optimization approach is introduced to select the optimal indicator combination. This involves considering all candidate indicator combinations... , and G Each index is normalized, and then unified according to the principle that larger values ​​are better, resulting in a multi-objective optimization evaluation index array. R noise , R boot and R G : (10) in, , and Representing the first g Group of candidate indicator combinations , and G The normalized value. For a , b Two sets of optimization evaluation indicators, if a Compared to b satisfy: (11) If at least one of them is strictly greater than, then it is called a group. a Dominant Group b All groups not dominated by other groups constitute the Pareto front, reflecting the optimal trade-off boundary of different candidate indicator combinations on the three optimal evaluation indicators. Furthermore, the final candidate indicator combination is selected in the Pareto front using mass loss constraints and the principle of closest ideal point; that is, the indicators in the combination must include mass loss to supplement the characterization of the severity of thermal runaway from the dimensions of mass release and structural damage. The ideal point is defined as the point where all three optimal evaluation indicators reach their optimum, denoted as: (12) However, in reality, such a combination is highly unlikely to exist. Therefore, let's consider Pareto front combinations that include quality loss. g The three optimization evaluation index vectors are z g : (13) Then its Euclidean distance to the ideal point L g for: (14) The final selection makes L gThe smallest combination is used as the comprehensive evaluation index for the severity of thermal runaway, and then the STRS quantitative value under this evaluation index is obtained.

[0036] Modeling the severity of thermal runaway: In this embodiment, Gaussian Process Regression (GPR) is used to establish a model for the severity of thermal runaway. It is a Bayesian nonparametric method suitable for small-sample, low-dimensional, nonlinear regression problems.

[0037] For the standardized input vector and observation output y = STRS , P h , SOC 0 and D C These represent heating power, initial SOC, and discharge rate, respectively, with the asterisk indicating standardization. The Gaussian regression process assumes: (15) in, f For regression function, i.e. x arrive y The mapping, Indicates difference from x This is used to describe any two distinct input vectors. For any input, GPR provides not only its predicted mean but also its corresponding predicted variance. m GP represents the Gaussian regression process, where GP is the mean function. k Let the covariance function be the covariance function, preferably the squared exponential kernel function, and its expression is: (16) in, It is the signal variance, which reflects the overall fluctuation range of the function. x i,d and x j,d Representing samples respectively i and j In the d Values ​​in each dimension d The values ​​1, 2, and 3 correspond to respectively P h , SOC 0 and D C . l Using length as a scale, it measures the difference between two samples. d Dimensional similarity is automatically estimated as a hyperparameter during training.

[0038] Security boundary construction: After training the GPR model, a dense input grid is constructed to extend the model from discrete experimental points to a continuous working condition space, which is then used as the input vector. x grid During the construction of the dense network, the step value of each parameter is set to Δ. P h Δ SOC 0 and Δ D C and generated respectively m , n , q indivual P h , SOC 0 and D C Inputting it into the trained GPR model returns the mean of the SRS predictions, which conform to a Gaussian distribution. u s and the predicted standard deviation : (17) in i The sample number. i ∈[1, m × n × q Predicted mean u s,i It can represent the average level of the predicted SRS distribution and can be used directly in most scenarios. u s,i Characterize and statistically analyze SRS. For example, you can set an SRS threshold. S th When the predicted mean u s,i Less than S th This can be considered sufficient to prevent thermal runaway. However, in the reverse calculation of the safety boundary for active thermal runaway suppression, the prediction uncertainty near the boundary needs special attention. Specifically, even the prediction mean... u s,i Less than S th It is still possible that within the confidence interval, [the following conditions] may still exist. u s,i + α ·σ s,i Greater than the safety threshold, among which α Let be the quantile of a cumulative probability in a standard normal distribution. Therefore, a one-sided confidence constraint needs to be introduced to construct a more conservative safety margin.

[0039] for us,i and σ s,i There is a 90% upper confidence threshold on one side. S U = u s,i + 1.2816·σ s,i The coefficient 1.2816 comes from the 90th quantile of the standard normal distribution. For a standard normal random variable... Z ~N(0,1), we have P ( Z ≤1.2816)≈0.90, which means that the true value has about a 90% probability of not exceeding the upper bound of the prediction. S U Therefore, the one-sided upper confidence boundary is... S U and S th Only by comparing S U Only when the safety threshold requirements are still met is the operating condition considered a safe zone. Therefore, the safe zone... A safe and safety boundary surface B safe It can be represented as: (18) (19) The above method incorporates model prediction uncertainty into boundary determination, thereby obtaining a more conservative and robust proactively suppressed safety boundary. This approach is more in line with engineering safety criteria, which dictates that it is better to shrink the safety zone than to misclassify uncertain conditions as safe conditions.

[0040] Active suppression strategies for thermal runaway After constructing the security boundary, for any P h , SOC 0 and D C In terms of operating conditions, first determine whether the area is within the safe zone based on the safety boundary, and then decide whether intervention is necessary. If the area is outside the safe zone, for P h , SOC 0 and D C As long as any operating condition variable is controllable, an active suppression space can be searched based on the safety boundary to determine the active suppression strategy. For example, within the controllable space, the classical gradient descent algorithm can be used to search for the operating condition variables corresponding to the SRS being within the safety boundary, and an appropriate safety margin can be reserved to cross the safety boundary and reach a deeper safety zone. Especially in specific... P h and SOCFind the discharge rate at 0 to reach the safe zone. D C Then, based on the obtained discharge rate D C Applying a discharge current (discharge current I=C) to the battery during thermal abuse. n D C C n (Based on the battery's rated capacity), it performs controlled discharge, thereby achieving active suppression of thermal runaway based on electro-regulation.

[0041] In this embodiment, the following tests were conducted: Commercial ternary lithium batteries with rated capacity of 5.2 Ah and charge / discharge cutoff voltages of 4.2 / 2.8 V were used. The tests covered three heating powers (200, 280, 360 W), three initial SOCs (40, 70, 100%), and four discharge rates (0, 1, 3, 5 C), constituting a total of 36 (3×3×4) test groups. Simultaneously, supplementary tests were conducted at a heating power of 280 W and a discharge rate of 0 C, with initial SOCs of 0%, 10%, and 20%, to analyze the thermal runaway characteristics at low SOCs. Ultimately, a total of 39 valid test groups were conducted. (Specifications are missing from the original text.) t TR_s and t TR_e These are the start and end times of thermal runaway, respectively, determined by the first time when the battery temperature increment is greater than or equal to 1 ℃ / s and less than or equal to -1 ℃ / s for 3 consecutive seconds.

[0042] Six candidate indicators were selected, namely, maximum temperature, maximum temperature rise rate, peak temperature rise increment, peak temperature drive ratio, excess temperature integral, and mass loss, denoted as Φ1 to Φ6.

[0043] 1) Φ1 Maximum temperature: The highest temperature of the battery during thermal runaway.

[0044] 2) Φ2 maximum temperature rise rate: the highest value of the battery temperature increase during thermal runaway.

[0045] 3) Peak temperature rise increment of Φ3: (20) in T For battery temperature, t For time. Φ3 is... T Maximum value minus the value at the start of thermal runaway T Value, equivalent to the period of thermal runaway T The increment. This indicator can mitigate the impact of differences in the initial temperature of battery thermal runaway under different operating conditions, focusing on the temperature rise caused by internal heat generation during thermal runaway, and correlated with the subsequent development intensity of thermal runaway.

[0046] 4) Peak temperature rise drive ratio of Φ4: (twenty one) in T h The temperature at which the heat source is triggered by heat abuse. ε It is a small positive number. Φ4 is Φ3 divided by the value at the start of thermal runaway. T h ( t TR_s )and T ( t TR_s The temperature difference. When Φ3 is the same, T h A higher value indicates a stronger external driving force and weaker internal heat generation, and vice versa. Therefore, this index can be regarded as the ability of thermal runaway to increase temperature under external thermal driving conditions.

[0047] 5) Φ5 excess temperature integral: (twenty two) During thermal runaway, compared to the battery thermal runaway initiation temperature T ( t TR_s ), T The incremental integral. This index reflects both the temperature rise magnitude and duration, effectively characterizing the cumulative intensity of the thermal runaway process through integration.

[0048] 6) Φ6 mass loss: the difference between the initial mass of the battery and the mass of the failed battery debris.

[0049] Calculate the weighted STRS for all indicator combinations based on formulas (1) to (9), and calculate the STRS for all indicator combinations. , and G The result is as follows Figure 2 As shown in the figure, the horizontal axis represents the number of indicators in the candidate indicator combination. For candidate indicators Φ1 to Φ6, their value range is [1, 6]. Therefore, when the number of indicators is 1, it corresponds to 6 points in the figure, and when the number of indicators is 6, it corresponds to only 1 point. It can be seen that there are obvious contradictions and trade-offs between different candidate indicator combinations in the three-dimensional evaluation results. Therefore, it is necessary to introduce the idea of ​​multi-objective optimization to select the best indicator combination.

[0050] Furthermore, for the 63 (2 6 -1) Candidate index combinations: Based on the non-dominance principle, 17 combinations located on the Pareto front were obtained, such as... Figure 3 As shown.

[0051] Using mass loss constraints and the principle of closest ideal point, the optimal combination is selected as (Φ1, Φ3, Φ6), which represents the highest temperature, the temperature rise increment at peak value, and the mass loss, with weights of 0.3328, 0.3774, and 0.2898, respectively. This yields the optimal SRS sequence, as shown below. Figure 4 As shown, the results are sorted in ascending order of STRS. The STRS values ​​of the test groups that did not experience thermal runaway ranged from 0.0006 to 0.0718, while the STRS values ​​of the test groups that experienced thermal runaway ranged from 0.1252 to 0.9334. Therefore, 0.125 was selected as the STRS value. S th .

[0052] Furthermore, leave-one-out cross-validation was used to train and validate the thermal runaway severity model. For all 39 experimental samples, the basic idea of ​​leave-one-out cross-validation is: each time, one sample is selected as the test sample, and the remaining 38 are used as training samples. This is repeated 39 times, ensuring that each sample serves exactly one test sample. After the leave-one-out prediction is completed, the predicted values ​​of all test samples are summarized, and the coefficient of determination is used. R 2 The accuracy of the model is evaluated using metrics such as RMSE and MAE.

[0053] In contrast, a multiple linear regression model was used to establish... P h , SOC 0 and D C Mapping relationship with STRS: (twenty three) in, β 0 to β 3. Regression coefficients were estimated using the least squares method. Training and validation were performed using the same leave-one-out training and test sets as the GPR model, and the same metrics were used to evaluate model accuracy. Figure 5(a) shows the prediction results of the GPR model under the leave-one-out method, with the linear regression model as a comparison. The predicted mean of GPR was used as its point prediction output, making it comparable to the prediction results of the linear regression model. It can be seen that the prediction results of GPR are significantly closer to the ideal fit and can provide a prediction distribution. As shown in Figures 5(b) and 5(c), the extreme value residuals of the linear regression model are -0.2922 and 0.2780, respectively. In contrast, the extreme value residuals of the GPR model are -0.1331 and 0.1478, respectively, significantly better than the linear regression model. Meanwhile, the linear regression model... R 2 The RMSE and MAE indices were 0.8285, 0.1186, and 0.0918, respectively, while the indices of the GPR model were 0.9595, 0.0576, and 0.0454, respectively, demonstrating excellent prediction accuracy.

[0054] Then, a dense input network is constructed, and the trained GPR model is used to continuously predict the STRS at each grid point. When constructing the input network, P h , SOC 0 and D C The ranges are set to [200, 360], [40, 100], and [0, 5] respectively, with a step value Δ. P h Δ SOC 0 and Δ D C Let's set them to 10, 5, and 0.5 respectively. Therefore... P h , SOC 0 and D C The lengths are 17, 13, and 11 respectively, generating a total of 2431 scattered points, thereby realizing an extended description of the working condition space.

[0055] Finally, P h , SOC 0、 D C Starting from the initial operating condition (320, 80, 0), setting the target SRS to 0.07 will automatically find the endpoint of the operating condition variable (260, 50, 3) that satisfies the target, which is the action target of the active suppression strategy, as shown in Figure 6(a). In Figure 6(a), each point represents an operating condition combination, and the color of the point is mapped to the SRS value, as shown in the color scale bar on the right. The safety boundary is represented by a light red plane. For operating condition combination points in the safety zone, they are set to a semi-transparent color, while the color of operating condition combination points in the non-safe zone is consistent with the SRS value.

[0056] However, situations where everything is completely controllable are rare in practice. When one operating variable is uncontrollable and the other is controllable, active suppression can be achieved in a two-dimensional space. As shown in Figure 6(b), in SOC 0 is fixed at 70%. P h and D C When controllable, with an initial operating condition starting point of (320,70,0), setting the target SRS to 0.1 will automatically search for the operating condition variable endpoint (230,70,4.5). When two operating condition variables are uncontrollable and one is controllable, intervention in thermal runaway can only be carried out in one controllable direction. That is, under specific thermal power and initial SOC, the discharge rate that can reach the safe zone is found, thereby realizing active suppression of thermal runaway based on electrical control. As shown in Figure 6(b), in P h and SOC When 0 is fixed at 280 W and 70% respectively, under the initial operating condition starting point (280,70,0), the discharge rate can be searched upward according to the direction indicated by the arrow. At this time, by adjusting the discharge rate to be higher than 4.5 C, active suppression of thermal runaway can be achieved.

[0057] In summary, this embodiment provides a novel approach to the active suppression of thermal runaway under thermal abuse and forms a complete theoretical system covering index construction, modeling and prediction, and strategy formulation, which has high practical value.

[0058] Example 2 An active suppression system for thermal runaway in lithium-ion batteries based on electro-regulation, comprising: The optimal index combination determination module is configured to apply weights to physical field parameters that reflect the severity of battery thermal runaway based on the index correlation weighting method, calculate the comprehensive score of thermal runaway severity of each index combination, quantitatively evaluate each index combination from three dimensions: sampling noise robustness, sample perturbation robustness, and severity discrimination ability, and select the index combination with the optimal comprehensive score of thermal runaway severity using a multi-objective optimization method. The thermal runaway severity prediction module is configured to take operating condition variables including thermal abuse power, battery SOC and discharge rate as input, and use a pre-trained thermal runaway severity prediction model based on Gaussian process regression to obtain the predicted mean and standard deviation of the comprehensive score of thermal runaway severity. The active suppression strategy search module is configured to, based on the predicted mean and predicted standard deviation, back-calculate the active suppression safety boundary according to the upper confidence bound, search for the operating condition variable corresponding to the comprehensive score of thermal runaway severity within the safety boundary, and determine it as the active suppression strategy action target.

[0059] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, 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 one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).

[0060] 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.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for actively suppressing thermal runaway in lithium-ion batteries based on electro-regulation, characterized in that, Includes the following steps: Based on the index correlation weighting method, weights are applied to the physical field parameters that reflect the severity of battery thermal runaway, and the comprehensive score of thermal runaway severity of each index combination is calculated. The index combination is quantitatively evaluated from three dimensions: sampling noise robustness, sample perturbation robustness, and severity discrimination ability. A multi-objective optimization method is used to select the index combination with the best comprehensive score of thermal runaway severity. Using operating condition variables including thermal abuse power, battery SOC and discharge rate as inputs, a pre-trained thermal runaway severity prediction model based on Gaussian process regression is used to obtain the mean and standard deviation of the comprehensive score prediction for the thermal runaway severity of the optimal combination of indicators. Based on the predicted mean and predicted standard deviation, the active suppression safety boundary is inferred from the upper confidence bound. The operating condition variable corresponding to the comprehensive score of thermal runaway severity within the safety boundary is searched and determined as the action target of the active suppression strategy.

2. The method for actively suppressing thermal runaway of lithium-ion batteries based on electro-regulation as described in claim 1, characterized in that, The process of assigning weights to physical field parameters reflecting the severity of battery thermal runaway based on the index correlation weighting method and calculating the comprehensive score of thermal runaway severity for each index combination includes: assigning weights to each index using the index correlation weighting method; for candidate indices in each test sample, calculating the normalized value and standard deviation of the candidate indices based on their original values ​​and the maximum and minimum values ​​of each index in all test samples; then calculating the correlation coefficient between two indices; calculating the information content of each index based on the correlation coefficient; calculating the objective weight; and finally calculating the comprehensive score vector of thermal runaway severity based on the objective weight. S : ； in, S i For the first i A comprehensive score reflecting the severity of thermal runaway for each test sample. w j For objective weighting, n The total number of test samples, m The total number of candidate indicators in each sample. The normalized value of the candidate indicator; for m There are 2 candidate indicators, which can generate a total of 2 m -1 candidate indicator combination, and for all candidate indicator combinations, calculate the comprehensive score of thermal runaway severity.

3. The method for actively suppressing thermal runaway of lithium-ion batteries based on electro-regulation as described in claim 1, characterized in that, The process of quantitatively evaluating the combination of indicators from the perspective of sampling noise robustness includes: calculating the weight of each indicator using the original data; then, for the normalized values ​​of all indicators, the repeated injection noise intensity is... σ Gaussian white noise is used. After each noise injection, a comprehensive score array of thermal runaway severity is calculated based on the noise data, and finally a comprehensive score matrix of thermal runaway severity is obtained. For each column, the coefficient of variation of each test sample under noise perturbation is calculated. The mean of all coefficients of variation is calculated as the sampling noise robustness evaluation index. The smaller the mean, the less sensitive the comprehensive score of thermal runaway severity obtained by the corresponding index combination is to index perturbation, and the stronger the noise robustness.

4. The method for actively suppressing thermal runaway of lithium-ion batteries based on electro-regulation as described in claim 1, characterized in that, The process of quantitatively evaluating the combination of indicators from the perspective of sample perturbation robustness includes: bootstrap resampling with replacement of the original experimental samples, with the same sample size. n The sample set is repeated multiple times to form multiple Bootstrap sample sets, each containing... n For each Bootstrap sample set, the weights of each indicator are recalculated and reapplied to the original experimental samples to calculate the comprehensive score array of thermal runaway severity. One row represents the comprehensive score array of thermal runaway severity for all experimental samples under a certain Bootstrap resampling weighting, and one column represents all experimental samples under Bootstrap resampling weighting. i The severity of thermal runaway is comprehensively scored. For each column, the coefficient of variation of each test sample under Bootstrap resampling weighting is calculated. The mean of all coefficients of variation is calculated as the sample perturbation robustness evaluation index. The smaller the value, the less sensitive the corresponding index combination is to sample sampling perturbation and the stronger the robustness under random sample set.

5. The method for actively suppressing thermal runaway of lithium-ion batteries based on electro-regulation as described in claim 1, characterized in that, The process of quantitatively evaluating each indicator combination from three dimensions of severity discrimination ability includes: using Gini average distance to quantify the discrimination between different sample scores of the comprehensive score of thermal runaway severity calculated from different indicator combinations. The larger the value, the greater the distance between values ​​in the SRS array, the greater the discrimination, and the more beneficial it is for subsequent modeling calculations.

6. The method for actively suppressing thermal runaway of lithium-ion batteries based on electro-regulation as described in claim 1, characterized in that, The process of selecting the optimal combination of indicators for the severity of thermal runaway using a multi-objective optimization method includes: normalizing the quantitative evaluation results of all candidate indicator combinations from three dimensions—sampling noise robustness, sample perturbation robustness, and severity discrimination ability—and then unifying them according to the principle that larger values ​​are better, thus obtaining a multi-objective optimized evaluation indicator array. a , b Two sets of optimization evaluation indicators, if a Group relative b Group satisfies a All quantitative evaluation results of the group were greater than or equal to b A group is defined as a group whose quantitative evaluation results are strictly greater than a certain threshold, and at least one of these results is greater than a certain threshold. a Dominant Group b All groups not dominated by other groups constitute the Pareto front. In the Pareto front, the final candidate index combination is selected by using the mass loss constraint and the principle of the closest ideal point.

7. The method for actively suppressing thermal runaway of lithium-ion batteries based on electro-regulation as described in claim 1, characterized in that, The thermal runaway severity prediction model based on Gaussian process regression takes thermal abuse power, battery SOC and discharge rate as inputs and the observed value of the comprehensive score of thermal runaway severity as output. The mapping relationship between input and output is represented by Gaussian regression process.

8. The method for actively suppressing thermal runaway of lithium-ion batteries based on electro-regulation as described in claim 1, characterized in that, The process of inferring the active suppression safety boundary based on the predicted mean and predicted standard deviation, according to the upper confidence bound, includes: based on the predicted mean... u s,i and the predicted standard deviation σ s,i By introducing a one-sided confidence constraint, a predetermined percentage upper confidence bound for the comprehensive score of thermal runaway severity is calculated. S U = u s,i + α ·σ s,i , α Let quantile be a cumulative probability quantile in a standard normal distribution, when the predicted mean is... S U Less than the preset threshold S th At that time, it was considered that thermal runaway could be avoided, and it was regarded as a safe zone, and then the safe boundary surface was determined.

9. The method for actively suppressing thermal runaway of lithium-ion batteries based on electro-regulation as described in claim 1, characterized in that, The process of determining the operating condition variables corresponding to the comprehensive score of the severity of thermal runaway within the safety boundary and identifying them as the action targets of the active suppression strategy includes: after constructing the safety boundary, for any operating condition of thermal abuse power, battery SOC, and discharge rate, firstly, it is determined whether it is in the safe zone based on the safety boundary, and then it is decided whether it is necessary to intervene. If it is outside the safe zone, for any of the operating condition variables of thermal abuse power, battery SOC, and discharge rate that are controllable, the active suppression space is searched based on the safety boundary, and the active suppression strategy is determined.

10. An active suppression system for thermal runaway in lithium-ion batteries based on electro-regulation, characterized in that, include: The optimal index combination determination module is configured to apply weights to physical field parameters that reflect the severity of battery thermal runaway based on the index correlation weighting method, calculate the comprehensive score of thermal runaway severity of each index combination, quantitatively evaluate each index combination from three dimensions: sampling noise robustness, sample perturbation robustness, and severity discrimination ability, and select the index combination with the optimal comprehensive score of thermal runaway severity using a multi-objective optimization method. The thermal runaway severity prediction module is configured to take operating condition variables including thermal abuse power, battery SOC and discharge rate as input, and use a pre-trained thermal runaway severity prediction model based on Gaussian process regression to obtain the mean and standard deviation of the comprehensive score prediction of thermal runaway severity with the optimal combination of indicators. The active suppression strategy search module is configured to, based on the predicted mean and predicted standard deviation, back-calculate the active suppression safety boundary according to the upper confidence bound, search for the operating condition variable corresponding to the comprehensive score of thermal runaway severity within the safety boundary, and determine it as the active suppression strategy action target.

Images