Lithium-ion Battery Thermal Safety Evaluation System and Method

Abstract

The present invention relates to the technical field of lithium batteries, and specifically, to a thermal safety evaluation system and method for lithium-ion batteries, which includes the following steps: Step 1: Collect multi-dimensional data during the operation of the lithium battery and process it as input parameters; Step 2: Determine the abuse type according to the input parameters; Step 3: Complete the evaluation of the thermal safety state of the battery using different thermal safety evaluation models according to different abuse types. The present invention can evaluate the safety state of the battery in real time by correlating multi-dimensional characteristic parameters of the lithium battery, classify the safety level of the battery using a model, and can also provide early warning information for the occurrence and evolution of the battery thermal runaway behavior.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium batteries, and more specifically, to a thermal safety evaluation system and method for lithium-ion batteries. Background Art

[0002] Lithium-ion batteries have been widely used in various fields, such as consumer electronics, automobiles, grid energy storage, etc. However, the thermal safety problem that accompanies its development has become a major obstacle to its popularization and application. The abuse of lithium-ion batteries under mechanical, electrical, and thermal environments may lead to serious thermal runaway. This runaway usually stems from side reactions inside the battery, resulting in heat accumulation, and ultimately triggering a series of irreversible heat generation processes, which may lead to serious consequences such as fire or explosion.

[0003] Currently, the evaluation techniques for the thermal safety of lithium-ion batteries are still relatively scarce. The existing evaluation methods mainly focus on the thermal safety testing of batteries or the risk probability assessment of thermal abuse, such as the invention patent: A method for analyzing the thermal safety of lithium batteries (CN 112098861 A). This patent uses a differential scanning calorimeter to test by analyzing the thermal safety characteristics of the internal structure of lithium batteries, and simply analyzes each component of the lithium battery, such as the positive electrode, negative electrode, and separator materials, and directly divides the dangerous deterioration threshold. Compared with this patent, there is a lack of a scientific and quantitative evaluation method, and it does not have the ability to update and optimize the method model in real time according to instant sample data. Another example is the invention patent: A method for evaluating the thermal abuse safety risk of a lithium battery pack based on multi-physical field simulation (CN 114692244A). This patent generates working conditions and battery sample data through mathematical methods, and then evaluates the risk probability of thermal abuse of lithium batteries by establishing safety critical values. Compared with this patent, these methods are based on the factory test data of batteries or the ideal working condition data generated by mathematical model simulation. However, due to factors such as battery aging, consistency, and various complex external working conditions, there may be a large difference between these test conditions and the actual use scenarios, resulting in the evaluation results being unable to accurately reflect the instant thermal safety of the battery, and the evaluation methods mainly rely on simple threshold judgments. In addition, the current risk probability assessment also lacks a multi-dimensional comprehensive analysis of the harm degree of lithium batteries under different abuse degrees, and cannot show the multi-dimensional comprehensive instant thermal safety of lithium batteries.

[0004] With the continuous progress of battery technology and the diversification of application scenarios, the requirements for battery thermal safety are also constantly increasing. Future research needs to introduce advanced sensing technologies and data analysis methods to better understand the chemical and physical changes inside the battery and provide more accurate safety evaluations. Improving the thermal safety of lithium-ion batteries not only requires improving existing materials and designs, but also strengthening the dynamic monitoring and evaluation of batteries in different environments. This will help ensure the safety and reliability of lithium-ion batteries in various applications, thereby promoting their wider application and development. Summary of the Invention

[0005] The content of the present invention is to provide a thermal safety evaluation system and method for lithium-ion batteries, aiming to solve the problem that the thermal safety evaluation of lithium-ion batteries is unclear, and the traditional method temporarily cannot solve the safety evaluation of lithium batteries at different stages under different abuse conditions and cannot distinguish its danger level.

[0006] A thermal safety evaluation method for lithium-ion batteries according to the present invention includes the following steps:

[0007] Step 1: Collect multi-dimensional data during the operation of the lithium battery, and process it as input parameters;

[0008] Step 2: Determine the abuse type according to the input parameters;

[0009] Step 3: Complete the thermal safety status evaluation by using different thermal safety evaluation models according to different abuse types.

[0010] Preferably, in Step 1, specifically:

[0011] 1.1) Deploy multi-parameter battery data acquisition sensors at key parts on the battery surface for data acquisition;

[0012] 1.2) Obtain the cycle aging data of the battery cell, including instantaneous voltage, current, temperature, pressure, capacity attenuation, and internal resistance;

[0013] 1.3) Process the data to obtain temperature-time change rate, pressure-time change rate, capacity retention rate, and voltage-time change rate data;

[0014] 1.4) Smoothly process the data by using the Kalman filter algorithm and analyze it;

[0015] 1.5) Use the processed data as input parameters.

[0016] Preferably, in Step 2, specifically:

[0017] 2.1) After receiving the input parameters, if T>60°C or dF / dt>25 kPa / s, it means that the battery is in an unsafe state, and the input parameters are combined with the empirical learning model for abuse analysis. Otherwise, it is considered that the battery is safe and no abuse analysis is performed; where T represents the battery temperature, F represents the battery expansion pressure, and t represents time;

[0018] 2.2) If dT / dt≥1°C / s, directly report to the warning system and jump out of the abuse judgment; if 0.02°C / s<dT / dt<1°C / s, match and judge the abuse type in the thermal abuse discrimination model, electrical abuse discrimination model, and mechanical abuse discrimination model;

[0019] 2.3) After the data is empirically and logically discriminated based on three discrimination models, it is transferred to the historical expert database for auxiliary discrimination. The data in the historical expert database comes from the changes in thermal safety data of historical same-type batteries under different abuse conditions. Based on the historical experimental data, weighted weights are assigned to the discrimination results respectively;

[0020] 2.4) Distinguish the logical judgment intervals of the three discrimination models, and select the abuse types corresponding to the corresponding machine learning training intervals as the judgment results;

[0021] Here, the softmax function is used to perform logistic regression division on the three abuse types;

[0022]

[0023] Among them, C k represents the category, K' is taken as 3, x represents the input feature vector, and Z k represents the linear combination of the feature vector x under the k-th category.

[0024] Preferably, in step 2.3), specifically:

[0025] In the electrical abuse discrimination model, if H2 is detected first and at the same time the voltage V is monitored to reach ±15% of the specified cut-off voltage of the battery, it is judged as electrical abuse; according to the increase or decay amount of the voltage, a hazard assignment of 0-1 is performed;

[0026] In the thermal abuse discrimination model, if the evolution of organic gas is detected first and the ambient temperature around the battery is above 100°C at the same time, it is judged as thermal abuse; according to the ratio of the organic gas to the CO concentration, the higher the proportion of the organic gas concentration, the greater the danger, and a hazard assignment of 0-1 is performed;

[0027] In the mechanical abuse discrimination model, if it is detected first that the reading of the pressure sensor is greater than 1 MPa due to an external force applied to the battery, and the time for the battery temperature to rise from the normal operating temperature to above 60°C is less than 5 s or the battery voltage drops to 50% or less of the cut-off discharge voltage, it is judged as mechanical abuse; according to the change degree of the dT / dt and dF / dt values, a hazard assignment of 0-1 is performed.

[0028] Preferably, in step 3, the thermal safety assessment model includes an electrical abuse model, a thermal abuse model, and a mechanical abuse model;

[0029] First, the correlation between the model input parameters and the predicted values is analyzed using the Pearson correlation coefficient, which ranges from [-1, 1]. The correlation strength references 0.8 - 1.0, 0.6 - 0.8, 0.4 - 0.6, 0.2 - 0.4, and 0.0 - 0.2 for highly correlated, strongly correlated, moderately correlated, weakly correlated, and extremely weakly correlated, respectively. The following formula is used for discrimination:

[0030]

[0031] ρ represents the correlation coefficient, x1 and x2 represent two different samples, cov is the covariance, Dx is the variance, indicating the degree of deviation between the variable and the expectation; Ex is the expectation, that is, the average value of the variable;

[0032] For the electrical abuse model, the voltage is a significant feature, and its model determination features include voltage, temperature, resistance, and gas, to evaluate the thermal safety state of lithium batteries;

[0033] First, each parameter is estimated as an independent estimator, and finally, the prediction results of each estimator are regressively integrated through a combiner; specifically, the final regression result F can be expressed as the average of the results of all estimators, that is:

[0034]

[0035] f i is the prediction result of the i-th estimator, and n is the number of estimators;

[0036] To train each estimator, sampling with replacement is performed from the operating or experimental data with a sample size of N, repeated N times, to form N training subsets; these samples are used to train independent estimators and become the training data for the estimator nodes;

[0037] In a sample with M attributes, when each estimator's node splits, m attributes are randomly selected from them, where m is much smaller than M; the node selects one of these m attributes as the best splitting attribute, and the splitting process continues until the model performance no longer improves after the node splits; a large number of estimators are constructed in this way to form a complete electrical abuse evaluation model;

[0038] For the thermal abuse model, the temperature change rate is a significant feature, and its model determination features include temperature, voltage, and gas composition, to evaluate the thermal safety state of lithium batteries;

[0039] After the estimator splitting is also performed, due to the data error of the thermal abuse model, an optimization algorithm needs to be used to tune the hyperparameters of the model to find the optimal combination, as follows:

[0040] K-fold cross-validation is used here. The dataset is divided into K equal parts, with one part as the validation set and the remaining K - 1 parts as the training set. According to this logic, repeat K times to obtain K models and their corresponding errors. Calculate the average of all errors to get the generalization error E of the model, as shown in the following formula:

[0041]

[0042] At the same time, a smoother covariance function is used as the covariance function for calculating the Gaussian process:

[0043]

[0044] σ 2 represents variance, and r represents the distance difference between signals;

[0045] For the mechanical abuse model, pressure is a significant feature. Its model determination features include pressure, voltage, and temperature, and are used to evaluate the thermal safety state of lithium batteries. The evaluation method is the same as above. Considering the special danger of mechanical abuse, if the battery temperature rise dT / dt is greater than 0.5 °C / s during the evaluation process, the warning system will be directly linked to report a level 3 alarm.

[0046] Preferably, in step 3, the evaluation method is:

[0047] Based on the results of three thermal safety evaluation models, the thermal safety state of lithium batteries is classified;

[0048] The evaluation safety interval is [0, 1]. Referring to the state of health (SOH) of the battery, when SOH < 0.8, the battery needs to be replaced. Similarly, define the value σ. When the evaluation safety value is in the interval [σ, 1], it is considered that the battery temperature rise is relatively safe and controllable under abuse. When the battery evaluation safety calculation value is in the interval [0, σ], it is considered that the battery temperature rises rapidly and has poor safety under different abuse conditions, and its thermal safety is uncontrollable;

[0049] In addition, the safety state level of lithium batteries is divided into four levels: safe, level 1 warning, level 2 warning, and level 3 warning. Set thresholds σ1, σ2, and σ3 according to different types of batteries;

[0050] Compare the interval where the evaluation value is located to conduct a physical examination rating of the battery safety. When the evaluation safety value is in the interval [0, σ1], it is safe. When the evaluation safety value is in the interval [σ1, σ2], it is a level 1 alarm. When the evaluation safety value is in the interval [σ2, σ3], it is a level 2 alarm. When the evaluation safety value is in the interval [σ3, 1], it is a level 3 alarm.

[0051] The present invention provides a thermal safety evaluation system for lithium-ion batteries, which adopts the above-mentioned thermal safety evaluation method for lithium-ion batteries.

[0052] Through the coupled analysis of battery aging data and thermal runaway data, the present invention establishes a thermal safety model to complete the quantitative evaluation of the thermal safety of lithium-ion batteries and ensure the safe use of the batteries. Through various relevant parameters, the present invention can evaluate the current safety of lithium batteries, and using the model, the batteries can be classified according to the safety level, which can provide early warning information for the occurrence and evolution of battery thermal runaway behavior. Description of the Drawings

[0053] Figure 1 It is a flowchart of a method for evaluating the thermal safety of a lithium-ion battery in an embodiment. Detailed Embodiments

[0054] To further understand the content of the present invention, the present invention will be described in detail in combination with the drawings and embodiments. It should be understood that the embodiments are only for explaining the present invention and not for limiting it.

[0055] Embodiment

[0056] As Figure 1 shown, this embodiment proposes a method for evaluating the thermal safety of a lithium-ion battery, and conducts a thermal safety evaluation of a commercial lithium titanate battery (rated capacity 9 Ah), which includes the following steps:

[0057] Step 1: Collect multi-dimensional data during the operation of the lithium battery, and process it as input parameters;

[0058] In step 1, specifically:

[0059] 1.1) Deploy multi-parameter battery data acquisition sensors at key parts on the battery surface for data acquisition;

[0060] 1.2) Obtain the cycle aging data of the battery cell, and the data is obtained through the individual charge and discharge tests of the battery; including instantaneous voltage, current, temperature, sensor pressure value, capacity at different aging cycles, resistance retention rate, etc.;

[0061] 1.3) Process the data to obtain data such as temperature-time change rate, pressure-time change rate, capacity retention rate, voltage-time change rate, etc.;

[0062] 1.4) Perform smoothing processing on the data using the Kalman filter algorithm to reduce errors and conduct analysis;

[0063] 1.5) Use the processed data as input parameters.

[0064] Step 2: Determine the abuse type according to the input parameters;

[0065] In step 2, specifically:

[0066] 2.1) After receiving the input parameters, the temperature around the battery is measured to be 118 °C, indicating that the battery is in an unsafe state. The input parameters are combined with the empirical learning model for abuse analysis;

[0067] 2.2) At this time, the analysis shows that dT / dt is 0.15 °C / s. Then, the abuse type is judged by matching in the thermal abuse discrimination model, electrical abuse discrimination model, and mechanical abuse discrimination model;

[0068] At this time, the sensor first detects the CO gas component, and the temperature around the battery is greater than 100 °C. The compliance of the three major abuse determination models is comprehensively matched. The results are as follows: the thermal abuse is 0.88, the electrical abuse is 0.10, and the mechanical abuse is 0.02. It is initially judged that thermal abuse has occurred, and the weight value is set to 0.7;

[0069] 2.3) The expert data and historical expert database are called for auxiliary discrimination. The data in the historical expert database comes from the aging data changes of the same type of battery under different abuse conditions in history. Based on the historical experimental data, the current model is weighted, and the weight value is 0.3;

[0070] 2.4) Distinguish the logical judgment intervals of the three discrimination models, and select the abuse type corresponding to the corresponding machine learning training interval as the judgment result;

[0071] Here, the softmax function is used to perform logical regression division on the three abuse types;

[0072]

[0073] Among them, C k represents the category, K' takes 3, x represents the input feature vector, and Z k represents the linear combination of the feature vector x under the k-th category.

[0074] Substitute the processed battery parameter information into the model, and the probability outputs are thermal abuse 0.85, electrical abuse 0.10, and mechanical abuse 0.05; after comprehensive weighting, it can be obtained that the thermal abuse is 0.871, the electrical abuse is 0.1, and the mechanical abuse is 0.029. It is judged that the system has encountered thermal abuse.

[0075] In step 3, since it is determined that the system has encountered thermal abuse, the thermal abuse assessment model is called;

[0076] First, calculate the correlation between the model input parameters and the safety evaluation value, and analyze it through the Pearson correlation coefficient. The value range is [-1, 1]. The correlation intensity references 0.8 - 1.0, 0.6 - 0.8, 0.4 - 0.6, 0.2 - 0.4, 0.0 - 0.2 as highly correlated, strongly correlated, moderately correlated, weakly correlated, and extremely weakly correlated respectively; the following formula is used for discrimination:

[0077]

[0078] ρ represents the correlation coefficient, x1 and x2 represent two different samples, cov is the covariance, Dx is the variance, indicating the degree of deviation between the variable and the expectation; Ex is the expectation, that is, the variable average value.

[0079] For the thermal abuse model of this example lithium titanate battery, the temperature change rate is a significant feature, and its model determination features include temperature, voltage, resistance, and gas composition to evaluate the thermal safety state of the lithium battery;

[0080] The temperature change rate, temperature, voltage, resistance, and CO concentration are respectively used as an independent estimator for estimation, and finally the prediction results of each estimator are regressively integrated through a combiner. Specifically, the final regression result F can be expressed as the average value of the results of all estimators, that is:

[0081]

[0082] f i is the prediction result of the i-th estimator, n is the number of estimators, and here n is 5;

[0083] First, train each estimator. We perform sampling with replacement from 100 sample data and repeat 100 times to form 100 training subsets. These samples are used to train independent estimators and become the training data of the estimator nodes to improve the model robustness.

[0084] In a sample with M attributes, when each estimator's node splits, m attributes are randomly selected from them, where m is much smaller than M. The node selects one of these m attributes as the best splitting attribute, and the splitting process continues until the model performance cannot be significantly improved after the node splits, specifically manifested as the improvement degree in this time being less than 1% compared to the previous time. Through this method, we can construct a large number of estimators, thus forming a complete abuse assessment model.

[0085] Then, use K-fold cross-validation. Divide the dataset into 4 equal parts, with one part as the validation set and the remaining 3 parts as the training set; according to this logic, repeat 4 times to obtain 4 models and the corresponding errors, and calculate the average value of all errors to obtain the generalization error E of the model, as shown in the following formula:

[0086]

[0087] At the same time, use a smoother covariance function formula as the covariance function for calculating the Gaussian process:

[0088]

[0089] σ 2 represents variance, and r represents the distance difference between signals;

[0090] Finally, we evaluate the performance of the model through the Mean Absolute Percentage Error (MAPE), Root Mean Square Percentage Error (RMSPE), and Coefficient of Determination (R 2 ).

[0091] Similarly, electro - abuse and mechanical - abuse analyses are carried out for the remaining cases; the results are shown in Table 1, which are the evaluation results of the electro - abuse model, mechanical - abuse model, and thermal - abuse model.

[0092] Table 1 Evaluation Results of Electro - abuse Model, Mechanical - abuse Model, and Thermal - abuse Model

[0093] Model Name MAPE (%) RMSPE (%) <![CDATA[R 2 > Electrical Abuse Model 3.336 3.880 0.712 Mechanical Abuse Model 2.998 3.364 0.776 Thermal Abuse Model 3.228 3.625 0.737

[0094] From the data in Table 1, it can be seen that for different models, R 2 is all above 0.712, MAPE is all below 3.336, and RMSPE is all below 3.880. This shows good predictability and evaluability.

[0095] Preferably, in step 3, the evaluation method is as follows:

[0096] Based on the results of the three thermal - safety evaluation models, the thermal safety of the lithium - ion battery is classified;

[0097] The evaluation safety interval is [0, 1]. Referring to the State of Health (SOH) of the battery, when SOH < 0.8, the battery needs to be replaced. Similarly, the value σ is defined. When the evaluation safety value is in the interval [σ, 1], it is considered that the battery temperature rise is relatively safe and controllable under extreme abuse. When the battery evaluation safety calculated value is on [0, σ], it is considered that the battery has a fast temperature rise and poor safety under different abuse conditions, and its thermal safety is uncontrollable;

[0098] In addition, the safety state level of the lithium - ion battery is divided into four levels: safe, first - level warning, second - level warning, and third - level warning. For the battery in this embodiment, the thresholds are set as σ1 = 0.25, σ2 = 0.58, and σ3 = 0.72;

[0099] According to the interval where the evaluation value 0.324 output by the model is located, at this time, the battery gives a first - level warning.

[0100] This embodiment provides a thermal - safety evaluation system for lithium - ion batteries, which adopts the above - mentioned thermal - safety evaluation method for lithium - ion batteries.

[0101] In this embodiment, a thermal safety model is established through the coupled analysis of battery aging data and thermal runaway data to complete the quantitative assessment of the thermal safety of lithium-ion batteries and ensure the safe use of the batteries. Through various relevant parameters in this embodiment, the current safety of lithium batteries can be evaluated, and using the model, the batteries can be classified according to the safety level, which can lay a foundation for the timely early warning of the future safety situation of the batteries.

[0102] The above schematically describes the present invention and its implementation manners. This description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Therefore, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments to this technical solution without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A method for thermal safety assessment of a lithium-ion battery, characterized in that: It includes the following steps: Step 1: Collect multi-dimensional data during the operation of the lithium battery, and use it as input parameters after processing; Step 2: Determine the type of abuse according to the input parameters; Step 3: Complete the thermal safety status assessment using different thermal safety assessment models according to different types of abuse; In Step 3, the thermal safety assessment model includes an electrical abuse model, a thermal abuse model, and a mechanical abuse model; First, the correlation between the model input parameters and the predicted values is analyzed through the Pearson correlation coefficient, and the value range is [-1, 1]. The correlation strength references 0.8-1.0, 0.6-0.8, 0.4-0.6, 0.2-0.4, and 0.0-0.2 as highly correlated, strongly correlated, moderately correlated, weakly correlated, and extremely weakly correlated, respectively. The following formula is used for discrimination: ρ represents the correlation coefficient, x1 and x2 represent two different samples, cov is the covariance, D() is the variance, indicating the degree of deviation between the variable and the expectation; E() is the expectation, that is, the variable average value; For the electrical abuse model, the voltage is a significant feature, and its model determination features include voltage, temperature, resistance, and gas, to evaluate the thermal safety status of the lithium battery; First, each parameter is used as an independent estimator for estimation, and finally, the prediction results of each estimator are regressively integrated through a combiner; specifically, the final regression result F is expressed as the average value of all estimator results, that is: f i is the prediction result of the i-th estimator, and n is the number of estimators; To train each estimator, sampling is performed with replacement from the operation or experimental data with a sample size of N, repeated N times, to form N training subsets; These samples are used to train independent estimators and become the training data of the estimator nodes; In a sample with M attributes, when each estimator's node splits, m attributes are randomly selected from them, where m is much smaller than M; The node selects one of these m attributes as the best splitting attribute, and the splitting process continues until the model performance no longer improves after the node splits; in this way, a large number of estimators are constructed to form a complete electrical abuse assessment model; For the thermal abuse model, the temperature change rate is a significant feature, and its model determination features include temperature, voltage, and gas composition, to evaluate the thermal safety status of the lithium battery; After the estimator split is also performed, due to the data error of the thermal abuse model, an optimization algorithm needs to be used to tune the hyperparameters of the model to find the optimal combination, specifically as follows: K-fold cross-validation is used here. The dataset is divided into K equal parts, one of which is used as the validation set, and the remaining K-1 parts are used as the training set; according to this logic, it is repeated K times to obtain K models and corresponding errors, and the average value of all errors is calculated to obtain the generalization error E of the model, as shown in the following formula: Meanwhile, a smoother covariance function formula is used as the covariance function for calculating the Gaussian process: σ 2 represents the variance, and r represents the distance difference between signals; For the mechanical abuse model, the pressure is a significant feature, and its model determination features include pressure, voltage, and temperature, to evaluate the thermal safety status of the lithium battery; considering the special danger of mechanical abuse, if the battery temperature rise dT / dt is greater than 0.5℃ / s during the evaluation process, the warning system will be directly linked to report a level 3 alarm.

2. The thermal safety assessment method for lithium-ion batteries according to claim 1, wherein: In Step 1, specifically: 1.1) Deploy multi-parameter battery data acquisition sensors at key parts of the battery surface for data acquisition; 1.2) Obtain the cyclic aging data of the battery cell, including the instantaneous voltage, current, temperature, pressure, capacity attenuation, and internal resistance; 1.3) Process the data to obtain the temperature-time change rate, pressure-time change rate, capacity retention rate, and voltage-time change rate data; 1.4) Smoothly process the data using the Kalman filtering algorithm and perform analysis; 1.5) Use the processed data as input parameters.

3. The lithium-ion battery thermal safety assessment method according to claim 2, wherein: In step 2, specifically: 2.1) After receiving the input parameters, if T > 60 °C or dF / dt > 25 kPa / s, it indicates that the battery is in an unsafe state. Combine the input parameters with the empirical learning model for abuse analysis. Otherwise, consider the battery safe and do not perform abuse analysis; where T represents the battery temperature, F represents the battery expansion pressure, and t represents time; 2.2) If dT / dt ≥ 1 °C / s, directly report to the warning system and jump out of the abuse judgment; if 0.02 °C / s < dT / dt < 1 °C / s, match and determine the abuse type in the thermal abuse discrimination model, electrical abuse discrimination model, and mechanical abuse discrimination model; 2.3) After performing empirical logical discrimination on the data based on the three discrimination models, call the historical expert database for auxiliary discrimination. The data in the historical expert database comes from the thermal safety data changes of historical same-type batteries under different abuse conditions. Based on the historical experimental data, assign weighted weights to the discrimination results respectively; 2.4) Distinguish the logical judgment intervals of the three discrimination models and select the abuse type corresponding to the corresponding machine learning training interval as the judgment result; Here, the softmax function is used to perform logistic regression division on the three abuse types; Among them, C k represents the category, K' takes 3, x represents the input feature vector, and Z k represents the linear combination of the feature vector x under the k-th category.

4. The method for evaluating the thermal safety of a lithium-ion battery according to claim 3, wherein: In step 2.3), specifically: In the electrical abuse discrimination model, if H2 is detected first and at the same time the voltage V is monitored to reach ±15% of the battery's specified cut-off voltage, it is judged as electrical abuse; assign a hazard value of 0-1 according to the increase or decrease of the voltage; In the thermal abuse discrimination model, if the evolution of organic gas is detected first and the ambient temperature around the battery is above 100 °C at the same time, it is judged as thermal abuse; assign a hazard value of 0-1 according to the ratio of the organic gas to the CO concentration. The higher the proportion of the organic gas concentration, the greater the danger; In the mechanical abuse discrimination model, if it is first detected that the pressure sensor reading is greater than 1 MPa due to an external force applied to the battery, and the time for the battery temperature to rise from the normal operating temperature to above 60 °C is less than 5 s or the battery voltage drops below 50% of the cut-off discharge voltage, it is judged as mechanical abuse; assign a hazard value of 0-1 according to the change degree of the dT / dt and dF / dt values; 5. The method for evaluating the thermal safety of a lithium-ion battery according to claim 4, wherein: In step 3, the evaluation method is: Grade the thermal safety state of the lithium battery based on the results of the three thermal safety evaluation models; The evaluation safety range is [0, 1]. Referring to the state of health (SOH) of the battery, when SOH < 0.8, the battery needs to be replaced. Similarly, a value σ is defined. When the evaluated safety value is in the range of [σ, 1], it is considered that the battery temperature rise under abuse is relatively safe and controllable. When the calculated value of the battery evaluation safety is on [0, σ], it is considered that the battery has a fast temperature rise and poor safety under different abuse conditions, and its thermal safety is uncontrollable. In addition, the safety state level of the lithium battery is divided into four levels: safe, first-level warning, second-level warning, and third-level warning, and thresholds σ1, σ2, and σ3 are set according to different types of batteries. By comparing the interval where the evaluation value is located, a physical examination rating of the battery safety is carried out. When the evaluated safety value is in the range of [0, σ1], it is safe. When the evaluated safety value is in the range of [σ1, σ2], it is a first-level alarm. When the evaluated safety value is in the range of [σ2, σ3], it is a second-level alarm. When the evaluated safety value is in the range of [σ3, 1], it is a third-level alarm.

6. Thermal safety assessment system for lithium-ion batteries, characterized in that: It adopts the lithium-ion battery thermal safety evaluation method described in any one of claims 1-5.

Images