Battery thermal runaway dynamic prediction method, system, device, medium and product

Abstract

The application discloses a battery thermal runaway dynamic prediction method, system, device, medium and product, and is characterized in that the method comprises the following steps: obtaining a plurality of operation data of a lithium ion battery; inputting the operation data into a first self-encoder for feature reconstruction to obtain first abnormal data, inputting the operation data into a second self-encoder for feature reconstruction to obtain second abnormal data, determining a first set, a second set and a target set based on the first abnormal data and the second abnormal data, respectively calculating the correlation between the data features in the first set, the second set and the target set, obtaining a thermal runaway early warning feature subset and a thermal runaway sensitive feature subset, and determining target operation data based on the thermal runaway early warning feature subset, the thermal runaway sensitive feature subset and the target set; and predicting the thermal runaway of the lithium ion battery based on the target operation data to obtain a thermal runaway prediction result of the lithium ion battery. The application can improve the accuracy of battery thermal runaway prediction.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and in particular to methods, systems, devices, media, and products for dynamic prediction of battery thermal runaway. Background Technology

[0002] With the widespread application of lithium-ion batteries in electric vehicles, large-scale energy storage systems, and portable electronic devices, battery safety issues under high energy density and complex operating conditions are becoming increasingly prominent. Under conditions such as overcharging, over-discharging, external short circuits, internal defects, or high-temperature environments, violent exothermic reactions may occur inside the battery, leading to thermal runaway and serious safety accidents such as fires or even explosions. Because battery thermal runaway is characterized by its suddenness, rapid evolution, and wide-ranging impact, passive protection after an accident often fails to prevent safety risks in a timely manner. Therefore, how to dynamically and proactively predict and warn of potential thermal runaway risks during battery operation has become a critical technical problem that urgently needs to be solved in battery management systems.

[0003] Existing battery thermal runaway prediction technologies mainly include physics-based and data-driven methods, but they still have significant shortcomings in prediction accuracy. Physics-based methods rely on precise modeling of the battery's electrochemical and thermal behavior, making it difficult to comprehensively characterize the nonlinear characteristics and individual differences of batteries under complex real-world operating conditions. The model parameters are sensitive and computationally complex, making the prediction results susceptible to model errors. While data-driven methods can learn features from historical data, they typically require a large amount of real thermal runaway sample data covering different operating conditions and fault modes. However, in practice, obtaining thermal runaway samples is costly and limited in quantity, resulting in insufficient model generalization ability and a tendency for misjudgments or omissions. Furthermore, existing methods often rely on a single anomaly detection mechanism, making it difficult to effectively distinguish between real anomalies and noise interference, thus affecting the accurate identification and reliable prediction of battery thermal runaway states. Summary of the Invention

[0004] This invention provides a method, system, device, medium, and product for dynamic prediction of battery thermal runaway, which can improve the accuracy of battery thermal runaway prediction.

[0005] In a first aspect, embodiments of the present invention provide a method for dynamic prediction of battery thermal runaway, comprising:

[0006] Obtain several operational data points for lithium-ion batteries;

[0007] Each of the aforementioned operating data is input into a first autoencoder for feature reconstruction to learn the electrothermal coupling change characteristics that deviate from normal operating conditions in each of the operating data, thereby obtaining first abnormal data. Each of the aforementioned operating data is input into a second autoencoder for feature reconstruction to learn the abnormal response characteristics in each of the operating data, thereby obtaining second abnormal data. Based on the feature distribution relationship between the first abnormal data and the second abnormal data, the correlation between each data feature is calculated, and based on the correlation, a thermal runaway early warning feature subset and a thermal runaway sensitive feature subset are selected respectively, thereby determining the target operating data for thermal runaway prediction.

[0008] Based on the target operating data, thermal runaway prediction of lithium-ion batteries is performed to obtain the thermal runaway prediction results of lithium-ion batteries.

[0009] This invention provides multi-dimensional and continuous foundational data support for subsequent thermal runaway feature mining by comprehensively reflecting the changes in the electrical and thermal states of the battery during actual operation, thereby improving the reliability of thermal runaway prediction. Starting from the normal operating baseline, it captures subtle coupling anomalies between parameters such as voltage, temperature, and internal resistance in the early stages of thermal runaway, allowing potential thermal runaway precursors to appear earlier, thus improving the foresight and sensitivity of thermal runaway prediction. By modeling the abrupt response and nonlinear amplification behavior in battery operating data from the perspective of anomaly patterns, it effectively characterizes the evolution of typical anomaly features during the development of thermal runaway, thereby enhancing the ability to identify real anomalies and reducing the probability of misjudgment. Through the differentiation of… This approach distinguishes between anomalous features appearing only from a single anomalous perspective and key anomalous features appearing together in multiple anomalous representations, providing a clear basis for subsequent feature selection and thus improving the accuracy of thermal runaway feature localization. By quantitatively screening key features highly correlated with thermal runaway evolution from a large number of anomalous features, the interference of redundant information and noisy features on the prediction model is reduced, thereby improving the effectiveness and discriminative power of thermal runaway prediction feature input. By realizing the fusion expression of multi-source and multi-stage anomalous information, the data input to the prediction stage contains both early warning information and reflects the anomalous evolution trend, thereby improving the completeness and accuracy of thermal runaway state representation as a whole, and significantly improving the accuracy and reliability of lithium-ion battery thermal runaway prediction results.

[0010] Furthermore, the step of inputting each of the operational data into the first autoencoder for feature reconstruction to learn the electrothermal coupling change characteristics deviating from normal operating conditions in each of the operational data, and obtaining the first abnormal data, includes:

[0011] Each of the aforementioned operational data is input into a first autoencoder for feature encoding to obtain first reconstructed data corresponding to each of the aforementioned operational data, wherein the first autoencoder is trained based on normal battery operating conditions;

[0012] Calculate the first reconstruction error between each of the said running data and each of the first reconstructed data;

[0013] The first reconstruction error is compared with a first preset threshold to filter each of the running data based on the first comparison result, thereby identifying the first abnormal data.

[0014] This invention reconstructs features by inputting battery operating data into a first autoencoder trained under normal operating conditions. The reconstruction error is used to characterize the deviation of the operating data from the normal electrothermal coupling characteristics. Combined with a preset threshold, the first abnormal data is obtained. This enables the effective identification of representative abnormal operating features in the early stages of thermal runaway, reducing the interference of noise data and instantaneous fluctuations. This provides high-quality input data for subsequent thermal runaway feature fusion and prediction models, thereby improving the accuracy and reliability of battery thermal runaway prediction.

[0015] Furthermore, the step of inputting each of the operational data into the second autoencoder for feature reconstruction to learn the abnormal response features in each of the operational data and obtain the second abnormal data includes:

[0016] Each of the aforementioned operational data is input into a second autoencoder for feature encoding to obtain second reconstructed data corresponding to each of the aforementioned operational data, wherein the second autoencoder is trained based on battery thermal runaway conditions;

[0017] Calculate the second reconstruction error between each of the said running data and each of the second reconstructed data;

[0018] The second reconstruction error is compared with a second preset threshold to filter each of the running data based on the second comparison result, thereby identifying the second abnormal data.

[0019] In this embodiment of the invention, battery operating data is input into a second autoencoder trained based on thermal runaway conditions for feature reconstruction. This allows the second autoencoder to model the abnormal response patterns exhibited by the battery during the thermal runaway evolution process. The reconstruction error characterizes the degree of matching between the operating data and typical thermal runaway features. Combined with a second preset threshold, second abnormal data is obtained, thereby highlighting the abnormal response features that are highly correlated with the thermal runaway evolution, and thus improving the accuracy and reliability of battery thermal runaway prediction.

[0020] Furthermore, the step of calculating the correlation between each data feature based on the feature distribution relationship between the first abnormal data and the second abnormal data includes:

[0021] The intersection of the first abnormal data and the second abnormal data is calculated to obtain the target set;

[0022] The data in the first abnormal data other than the target set is determined as the first set, and the data in the second abnormal data other than the target set is determined as the second set;

[0023] The data in the first set and the data in the second set are respectively compared with the data in the target set to calculate the feature correlation, and the first correlation result and the second correlation result are obtained respectively.

[0024] This invention enables the target set to centrally characterize key abnormal features highly correlated with battery thermal runaway by performing an intersection operation on the first abnormal data and the second abnormal data. At the same time, data that are identified as abnormal only in a single model are divided into the first set and the second set, reducing the risk of misjudgment caused by a single discrimination mechanism. This provides a clear data structure foundation for subsequent feature screening and fusion based on correlation, thereby improving the accuracy and stability of battery thermal runaway prediction.

[0025] Furthermore, the step of filtering out the thermal runaway early warning feature subset and the thermal runaway sensitive feature subset based on the correlation includes:

[0026] Based on the first correlation result, the data in the first set is filtered to obtain a subset of thermal runaway early warning features;

[0027] Based on the second correlation result, the data in the second set are filtered to obtain a subset of thermal runaway sensitive features.

[0028] This invention calculates the correlation between data features in the first and second sets and key anomalous features in the target set, enabling data identified as anomalous in a single model to be re-screened based on their correlation with core anomalous features of thermal runaway. This provides more discriminative and temporally consistent feature inputs for subsequent thermal runaway prediction models, thereby improving the accuracy and reliability of battery thermal runaway prediction.

[0029] Furthermore, the step of predicting thermal runaway of the lithium-ion battery based on the target operating data to obtain the thermal runaway prediction result of the lithium-ion battery includes:

[0030] The target operating data is subjected to feature normalization processing to obtain an abnormal feature sequence, wherein the target operating data is all data in the thermal runaway early warning feature subset, the thermal runaway sensitive feature subset, and the target set;

[0031] The abnormal feature sequence is input into a preset thermal runaway prediction model for inference and calculation to obtain a predicted value of thermal runaway risk.

[0032] The predicted thermal runaway risk value is compared with a third preset threshold to determine the thermal runaway risk level of the lithium-ion battery based on the third comparison result.

[0033] This invention normalizes the target operating data and inputs it into a thermal runaway prediction model for inference calculation, thereby achieving quantitative prediction of the thermal runaway risk of lithium-ion batteries. The corresponding risk level is determined by threshold comparison, thus improving the accuracy, stability, and engineering applicability of the thermal runaway prediction results.

[0034] Secondly, embodiments of the present invention provide a dynamic prediction system for battery thermal runaway, the system comprising: an acquisition module, a reconstruction module, and a prediction module;

[0035] The acquisition module is used to acquire several operational data of the lithium-ion battery;

[0036] The reconstruction module is used to input each of the operating data into a first autoencoder for feature reconstruction to learn the electrothermal coupling change characteristics that deviate from normal operating conditions in each of the operating data to obtain first abnormal data; input each of the operating data into a second autoencoder for feature reconstruction to learn the abnormal response characteristics in each of the operating data to obtain second abnormal data; calculate the correlation between each data feature based on the feature distribution relationship between the first abnormal data and the second abnormal data; and filter out thermal runaway early warning feature subset and thermal runaway sensitive feature subset based on the correlation to determine the target operating data for thermal runaway prediction.

[0037] The prediction module is used to predict the thermal runaway of the lithium-ion battery based on the target operating data, and obtain the thermal runaway prediction result of the lithium-ion battery.

[0038] This invention, through the establishment of an acquisition module, a reconstruction module, and a prediction module, utilizes a first autoencoder and a second autoencoder to reconstruct features from two perspectives: normal operating conditions and thermal runaway operating conditions. By combining feature distribution relationships and correlation analysis, key features with early warning significance and sensitivity are selected, thereby constructing target operating data for thermal runaway prediction. This enables accurate and stable prediction of the thermal runaway risk of lithium-ion batteries, improving the system's prediction reliability and engineering application value under complex operating conditions.

[0039] Thirdly, embodiments of the present invention provide a terminal device, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0040] The memory is used to store at least one executable instruction that causes the processor to perform the operation of the battery thermal runaway dynamic prediction method as described in this application.

[0041] Fourthly, embodiments of the present invention provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device or system where the computer-readable storage medium is located to perform the battery thermal runaway dynamic prediction method as described in this application.

[0042] Based on the above-described method embodiments, another embodiment of the present invention provides a computer program product, including a computer program or instructions, which, when executed by a communication device, implements the battery thermal runaway dynamic prediction method of any embodiment of the present invention.

[0043] The above description is merely an overview of the technical solutions of the embodiments of the present invention. In order to better understand the technical means of the embodiments of the present invention and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0044] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating an embodiment of the battery thermal runaway dynamic prediction method provided in this application;

[0046] Figure 2 This is a flowchart illustrating steps S201 to S203 provided in this application;

[0047] Figure 3 This is a flowchart illustrating steps S301 to S303 provided in this application;

[0048] Figure 4 This is a flowchart illustrating steps S401 to S402 provided in this application;

[0049] Figure 5 This is a schematic diagram of an embodiment of the battery thermal runaway dynamic prediction method provided in this application. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] Unless otherwise defined, 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 application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0052] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0053] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0054] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0055] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0056] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0057] With the widespread application of lithium-ion batteries in electric vehicles, large-scale energy storage systems, and portable electronic devices, their safety risks under high energy density and complex operating conditions are becoming increasingly prominent. Under conditions such as overcharging, over-discharging, short circuits, internal defects, or high-temperature environments, violent exothermic reactions may occur inside the battery, leading to thermal runaway and serious accidents such as fires or even explosions. Due to the suddenness and rapid evolution of battery thermal runaway, relying solely on post-accident protection is insufficient to avoid risks in a timely manner. Therefore, dynamic prediction and early warning of thermal runaway risks during operation are urgently needed. Existing thermal runaway prediction methods mainly include physical model-based and data-driven approaches. Among them, physical model methods are difficult to accurately characterize the nonlinear characteristics under complex operating conditions and have high computational complexity; data-driven methods are limited by the scarcity of thermal runaway samples and a single anomaly discrimination mechanism, which is prone to misjudgment or omission, resulting in insufficient prediction accuracy and reliability.

[0058] See Figure 1 To improve the accuracy of battery thermal runaway prediction, an embodiment of the present invention provides a dynamic prediction method for battery thermal runaway, including steps S101 to S103.

[0059] Step S101: Obtain several operational data points of the lithium-ion battery;

[0060] In some embodiments, acquiring several operational data points of the lithium-ion battery includes real-time acquisition of multi-dimensional state parameters of the target battery during operation via sensor modules, acquisition modules, and communication modules in the Battery Management System (BMS), to form a raw operational dataset for subsequent anomaly identification and thermal runaway prediction. The operational data includes at least one or more of electrical parameters, thermal parameters, and characteristic parameters reflecting changes in the battery's internal state. The electrical parameters include terminal voltage, current, internal resistance, voltage change rate, and current fluctuation amplitude of individual battery cells or battery modules. The thermal parameters include battery surface temperature, estimated internal temperature, temperature gradient, and temperature change rate. The characteristic parameters reflecting changes in the battery's internal state include state of charge (SOC), state of health (SOH), cell expansion, strain change, gas generation characteristic parameters, and gas concentration data such as CO, CO2, and H2 obtained from gas sensors. The operational data is continuously acquired according to a preset sampling period, which can be 10ms, 100ms, 1s, or adaptively set according to the battery type and application scenario to ensure that the operational data can fully reflect the dynamic changes of the battery under different operating conditions.

[0061] In some embodiments, after acquiring the operational data, a preprocessing step is further included. This preprocessing includes, but is not limited to, data time alignment, compensation for missing values, noise filtering, and normalization. The normalization process maps operational data with different dimensions to a unified numerical range to reduce the impact of differences in parameter scales on the subsequent autoencoder feature learning process. The operational data can be organized according to time windows to construct temporal data segments containing multiple consecutive sampling points, serving as the basic data units for subsequent input to normal and abnormal autoencoders, thereby enabling the model to learn the temporal evolution characteristics of battery operating states.

[0062] Through the above steps, comprehensive and continuous multi-dimensional operational data of lithium-ion batteries can be obtained. By preprocessing to eliminate noise, compensate for missing values, and normalize different parameter scales, high-quality time-series data segments can be constructed, providing a reliable data foundation for subsequent feature learning of autoencoders, thereby improving the accuracy and real-time performance of battery anomaly identification and thermal runaway prediction.

[0063] Step S102: Input each of the operating data into the first autoencoder for feature reconstruction to learn the electrothermal coupling change characteristics that deviate from normal operating conditions in each of the operating data to obtain the first abnormal data. Input each of the operating data into the second autoencoder for feature reconstruction to learn the abnormal response characteristics in each of the operating data to obtain the second abnormal data. Based on the feature distribution relationship between the first abnormal data and the second abnormal data, calculate the correlation between each data feature. Based on the correlation, filter out the thermal runaway early warning feature subset and the thermal runaway sensitive feature subset respectively, thereby determining the target operating data for thermal runaway prediction.

[0064] Please refer to Figure 2 In some embodiments, the step of inputting each of the operating data into a first autoencoder for feature reconstruction to learn the electrothermal coupling change characteristics in each of the operating data that deviate from normal operating conditions, and obtaining the first abnormal data, includes: steps S201 to S203.

[0065] Step S201: Input each of the said operating data into the first autoencoder for feature encoding to obtain the first reconstructed data corresponding to each of the said operating data, wherein the first autoencoder is trained based on the normal operating conditions of the battery;

[0066] In some embodiments, multi-source real-time operating data of the target battery during operation is input to a first autoencoder for feature encoding and decoding to obtain first reconstructed data corresponding one-to-one with each of the operating data. The operating data may include parameters reflecting the battery's electro-thermal coupling state, such as voltage, current, surface temperature, internal resistance, state of charge (SOC), temperature change rate, temperature difference, and gas generation characteristics. The first autoencoder is a normal autoencoder, which is trained using only historical operating data collected under normal battery operating conditions during the training phase. The training objective is to minimize the reconstruction error between the input data and the reconstructed data, thereby enabling the first autoencoder to learn and characterize the distribution pattern of electro-thermal coupling features under normal battery operating conditions.

[0067] In some embodiments, the first autoencoder includes an encoder and a decoder. The encoder maps high-dimensional operating data to a low-dimensional latent feature space to extract core features characterizing the normal operating condition of the battery. The decoder reconstructs the input operating data based on the latent features and outputs first reconstructed data with the same dimension as the input operating data. When the real-time operating data conforms to the normal operating condition distribution, the first autoencoder can reconstruct it with high accuracy; when the operating data deviates from the normal operating condition, its reconstruction capability will decrease significantly. The specific hierarchical structure and parameter configuration of the first autoencoder are shown in Table 1.

[0068] Table 1. Specific hierarchical structure and parameter configuration of the first autoencoder.

[0069]

[0070] Step S202: Calculate the first reconstruction error between each of the running data and each of the first reconstruction data;

[0071] In some embodiments, the first reconstruction error can be obtained by calculating the difference between each of the running data and the corresponding first reconstruction data. Specifically, it can be calculated using methods such as Euclidean distance, absolute error, mean square error, or weighted error. The calculation method is as follows:

[0072]

[0073] Among them, X i This represents the real-time running data of the i-th item. E represents the first reconstructed data output by the first autoencoder for the i-th running data. i The first reconstruction error is represented by ‖·‖, which represents the norm. The norm can be the Euclidean norm, a weighted Euclidean norm, or an absolute error sum, and can be specifically set according to the data type of the running data. In this embodiment, the first reconstruction error can be calculated for each piece of running data, thereby obtaining a sequence of first reconstruction errors corresponding to each piece of running data. The magnitude of the first reconstruction error is used to characterize the degree of deviation of the corresponding running data from the normal operating distribution. The larger the first reconstruction error, the more likely the running data is to not conform to the characteristics of the normal operating state.

[0074] Step S203: Compare the first reconstruction error with a first preset threshold to filter each of the running data based on the first comparison result and determine the first abnormal data.

[0075] In some embodiments, each first reconstruction error obtained in step S202 is compared with a first preset threshold one by one, and the real-time operating data of the target battery is filtered according to the comparison results to determine the first abnormal data. In this embodiment, the first preset threshold can be preset according to the statistical characteristics of reconstruction errors under normal battery operating conditions, for example, determined based on the mean and standard deviation of reconstruction errors of normal samples, or obtained by adaptive updating through historical operating data. Specifically, if the first reconstruction error corresponding to a certain real-time operating data is greater than the first preset threshold, it indicates that the operating data deviates significantly from the normal operating characteristic distribution learned by the first autoencoder, and the real-time operating data is judged as the first abnormal data; if the first reconstruction error corresponding to a certain real-time operating data is less than or equal to the first preset threshold, it is considered that the operating data is still within the normal operating range and is not regarded as the first abnormal data.

[0076] Please refer to Figure 3In some embodiments, the step of inputting each of the running data into a second autoencoder for feature reconstruction to learn the abnormal response features in each of the running data and obtain the second abnormal data includes: steps S301 to S303.

[0077] Step S301: Input each of the said operating data into the second autoencoder for feature encoding to obtain the second reconstructed data corresponding to each of the said operating data, wherein the second autoencoder is trained based on the battery thermal runaway condition;

[0078] In some embodiments, real-time operational data of the target battery during operation are input to a second autoencoder for feature encoding and decoding to obtain second reconstructed data corresponding one-to-one with each operational data point. The operational data may include, but is not limited to, voltage, current, surface temperature, temperature change rate, internal resistance, state of charge (SOC), pressure difference, temperature difference, gas concentration change, and derived feature data reflecting the degree of abnormal battery response. The second autoencoder is an anomaly autoencoder, which is trained using operational data under thermal runaway conditions fitted based on a battery physical model during the training phase. The training objective is to minimize the reconstruction error between the input data and the reconstructed data, thereby enabling the second autoencoder to learn and characterize the typical operational characteristic distribution of the battery under thermal runaway conditions.

[0079] In some embodiments, the second autoencoder includes an encoder and a decoder. The encoder maps high-dimensional operating data to a low-dimensional latent feature space to extract core abnormal features characterizing the thermal runaway state of the battery. The decoder reconstructs the input operating data based on the latent features and outputs second reconstructed data with the same dimension as the input operating data. Since the second autoencoder has been trained for thermal runaway conditions, its reconstruction capability is strong when the real-time operating data matches the distribution of thermal runaway state features; when the real-time operating data does not belong to the thermal runaway evolution features, its reconstruction error will increase significantly. The specific hierarchical structure and parameter configuration of the second autoencoder are shown in Table 2.

[0080] Table 2. Specific hierarchical structure and parameter configuration of the second autoencoder.

[0081]

[0082] Step S302: Calculate the second reconstruction error between each of the running data and each of the second reconstruction data;

[0083] In some embodiments, the difference between each set of operating data obtained in step S301 and the corresponding second reconstructed data is calculated to obtain the second reconstruction error corresponding to each set of operating data, so as to quantify the degree of matching between the real-time operating data and the characteristic distribution of the thermal runaway condition. The second reconstruction error can be calculated in the following way:

[0084]

[0085] Among them, X i This represents the i-th item of real-time running data. This represents the second reconstructed data output by the second autoencoder for the i-th running data. The second reconstruction error is represented by ||·||, which represents the norm. The norm can be the Euclidean norm, the weighted Euclidean norm, or the absolute error sum. In this embodiment, the weights of each dimension error can be set according to the importance of different operating parameters to improve the sensitivity to anomalies in key parameters such as temperature, voltage, or gas production.

[0086] Step S303: Compare the second reconstruction error with the second preset threshold to filter each of the running data based on the second comparison result and determine the second abnormal data.

[0087] In some embodiments, each second reconstruction error obtained in step S302 is compared with a second preset threshold, and the real-time operating data of the target battery is filtered according to the comparison results to determine the second abnormal data. In this embodiment, the second preset threshold can be preset according to the statistical characteristics of the reconstruction error of the second autoencoder on the thermal runaway training samples, for example, determined based on the mean, variance or quantile of the reconstruction error, or dynamically adjusted in combination with historical operating data. Specifically, when the second reconstruction error corresponding to a certain real-time operating data is less than or equal to the second preset threshold, it indicates that the operating data highly matches the thermal runaway condition feature distribution learned by the second autoencoder, and the real-time operating data is judged as a second abnormal data; when the second reconstruction error corresponding to a certain real-time operating data is greater than the second preset threshold, it indicates that the operating data does not conform to the thermal runaway condition feature distribution, and it is excluded from the second abnormal data.

[0088] In some embodiments, calculating the correlation between data features based on the feature distribution relationship between the first abnormal data and the second abnormal data includes: calculating the intersection of the first abnormal data and the second abnormal data to obtain the target set; determining the data in the first abnormal data other than the target set as the first set, and determining the data in the second abnormal data other than the target set as the second set; calculating the feature correlation between the data in the first set and the data in the second set and the data in the target set, respectively, to obtain a first correlation result and a second correlation result.

[0089] In some embodiments, the intersection of the first abnormal data and the second abnormal data is calculated to obtain the target set. Specifically, it is assumed that the real-time operating data of the target battery during a certain period includes multiple characteristic data such as voltage, current, temperature, internal resistance, and gas production. After the above steps, the first abnormal data set is A = {V1, T2, I3, R4, G5}, and the second abnormal data set is B = {T2, I3, T6, G7}. The intersection of the first abnormal data set A and the second abnormal data set B is calculated: intersection C = A ∩ B = {T2, I3}; where T2 represents temperature abnormal data number 2, and I3 represents current abnormal data number 3. This intersection C is taken as the target set, i.e., target set = {T2, I3}, and used for subsequent thermal runaway prediction data processing.

[0090] In some embodiments, the data in the first abnormal data excluding the target set is determined as the first set, and the data in the second abnormal data excluding the target set is determined as the second set. Specifically, according to the above example, the first abnormal data set A = {V1, T2, I3, R4, G5}, and the target set C = {T2, I3}, then the data in A excluding the target set is determined as the first set D1, i.e., D1 = {V1, R4, G5}; the second abnormal data set B = {T2, I3, T6, G7}, and the target set C = {T2, I3}, then the data in B excluding the target set is determined as the second set D2, i.e., D2 = {T6, G7}. In subsequent implementations, the correlation between each data item in the first set D1 and the second set D2 and the target set C can be calculated and filtered. The filtered candidate data is then merged with the target set C to obtain the final target abnormal data set for thermal runaway prediction.

[0091] In some embodiments, the data in the first set and the data in the second set are respectively compared with the data in the target set to calculate feature correlation, thereby obtaining a first correlation result and a second correlation result. Specifically, firstly, based on the first set and the target set obtained after the above steps, a correlation calculation model based on feature dimensions and time dimensions is constructed. Feature correlation analysis is then performed on the abnormal operation data in the first set and the abnormal operation data in the target set to quantify the degree of association between the data in the first set and the confirmed abnormal patterns. Specifically, let the first set be represented as... The target set is represented as Where x is a multidimensional feature vector composed of operating parameters such as voltage, current, temperature, rate of temperature change, internal resistance, and gas concentration. In this embodiment, the first correlation result is calculated using the Pearson correlation coefficient, and its calculation formula is as follows:

[0092]

[0093] Where, r ij Let K represent the first correlation result between the i-th data item in the first set and the j-th data item in the target set, where K is the number of feature dimensions. Next, based on the second set and the target set obtained after the above steps, feature correlation calculation is further performed to evaluate the degree of matching between the data in the second set and typical thermal runaway anomaly modes. Let the second set be denoted as... In this embodiment, the second correlation result is also calculated based on the Pearson correlation coefficient, and its calculation formula is consistent with that of the Pearson correlation coefficient, only the first set is replaced by the second set. Since the second set comes from the anomalous autoencoder, its data distribution is closer to the thermal runaway state. Therefore, the second correlation result is usually used to characterize the severity of the anomalous situation and the consistency of its evolution. In this embodiment, a weighting coefficient can also be introduced into the correlation calculation to assign higher weights to features that are more sensitive to thermal runaway, such as temperature and gas concentration, in order to improve the ability of the second correlation result to characterize key runaway features.

[0094] Please refer to Figure 4 In some embodiments, the step of calculating the correlation between each data feature in the first set, the second set and the target set respectively, and filtering out thermal runaway early warning feature subset and thermal runaway sensitive feature subset based on the correlation, includes: steps S401 to S402;

[0095] Step S401: Based on the first correlation result, filter the data in the first set to obtain a subset of thermal runaway early warning features;

[0096] In some embodiments, based on the first correlation result obtained in the foregoing steps, the abnormal operation data in the first set are further filtered to remove noisy data with low correlation to thermal runaway anomalies. In this embodiment, a first correlation threshold T1 is preset. When the first correlation result corresponding to a certain data in the first set and at least one data in the target set satisfies r ij If the value is ≥T1, then the data is considered to have a strong correlation with the abnormal characteristics of thermal runaway and is retained. After screening, a subset of thermal runaway early warning features is obtained, denoted as: This thermal runaway early warning feature subset mainly includes data that are in the early stages of thermal runaway evolution, have not yet fully conformed to the abnormal distribution of thermal runaway, but whose feature change trends are highly correlated with the target set, and are used to capture potential risk signals in advance.

[0097] Step S402: Based on the second correlation result, the data in the second set is filtered to obtain a subset of thermal runaway sensitive features.

[0098] In some embodiments, based on the second correlation result obtained in the foregoing steps, the abnormal operating data in the second set are filtered to extract the abnormal feature data most sensitive to thermal runaway. In this embodiment, a second correlation threshold T2 is set. When the second correlation result corresponding to a certain data in the second set and at least one data in the target set satisfies r... ij When the value is ≥T2, the data is classified as highly thermal runaway-related and included in the thermal runaway sensitive feature subset. The final thermal runaway sensitive feature subset is represented as follows: This subset of sensitive features for thermal runaway mainly reflects key abnormal characteristics of the battery that has entered the rapid evolution stage of thermal runaway. It can be used together with the subset of features for thermal runaway warning and the target set as core input data for subsequent thermal runaway prediction and risk classification.

[0099] Through the above steps, normal autoencoders and abnormal autoencoders can be used to extract early abnormal features of the battery deviating from normal operating conditions and typical abnormal features of thermal runaway, respectively. By analyzing the intersection of sets and the correlation of features, subsets of thermal runaway early warning features and subsets of thermal runaway sensitive features can be effectively screened, realizing a hierarchical characterization of the battery's abnormal state and providing core feature data with high correlation and high sensitivity for subsequent accurate prediction of thermal runaway risk.

[0100] Step S103: Based on the target operating data, perform thermal runaway prediction on the lithium-ion battery to obtain the thermal runaway prediction result of the lithium-ion battery.

[0101] In some embodiments, the step of predicting thermal runaway of a lithium-ion battery based on the target operating data to obtain a thermal runaway prediction result for the lithium-ion battery includes: performing feature normalization processing on the target operating data to obtain an abnormal feature sequence; inputting the abnormal feature sequence into a preset thermal runaway prediction model for inference calculation to obtain a predicted thermal runaway risk value; and comparing the predicted thermal runaway risk value with a third preset threshold to determine the thermal runaway risk level of the lithium-ion battery based on the third comparison result.

[0102] In some embodiments, the target operating data is subjected to feature normalization processing to obtain an abnormal feature sequence. The target operating data comprises all data from the thermal runaway early warning feature subset, the thermal runaway sensitive feature subset, and the target set. Specifically, before performing thermal runaway prediction based on the target abnormal operating data, feature normalization processing is first performed on the target abnormal operating data to eliminate the influence of differences in the dimensions, value ranges, and variation amplitudes of different operating parameters on the subsequent prediction model inference results, thereby improving prediction stability and accuracy. Specifically, the target abnormal operating data can be represented as: X = {x1, x2, ..., x...} T}, where x t =[x t,1 ,x t,2 ,…,x t,M Let be the multidimensional abnormal operation feature vector corresponding to the t-th sampling time, and M represent the number of abnormal feature dimensions. The abnormal features include, but are not limited to, abnormal voltage features, abnormal current features, abnormal temperature features, temperature change rate features, internal resistance change features, and gas production change features. In this embodiment, the target abnormal operation data is processed using a minimum-maximum normalization method, and the normalization formula is as follows:

[0103]

[0104] in, and Let these represent the minimum and maximum values ​​of the m-th dimension anomaly feature in historical anomaly samples or training data, respectively. These are the normalized anomaly feature values. Through the above normalization process, a sequence of anomaly features arranged in chronological order can be obtained: This abnormal feature sequence can more accurately reflect the dynamic evolution trend of the target battery under abnormal conditions, providing high-quality input features with a unified scale and comparability for subsequent thermal runaway prediction models.

[0105] In some embodiments, the abnormal feature sequence is input into a preset thermal runaway prediction model for inference calculation to obtain a predicted value of thermal runaway risk. Specifically, after the construction of the abnormal feature sequence is completed, the abnormal feature sequence is input into the preset thermal runaway prediction model for inference calculation to obtain a predicted value of the risk of thermal runaway of the target battery within a future time window. In this embodiment, the thermal runaway prediction model is a deep learning-based temporal prediction model, which includes a temporal feature extraction network and a risk mapping layer. The temporal feature extraction network is used to learn the temporal dependencies and evolutionary patterns in the abnormal feature sequence, and the risk mapping layer is used to map the extracted high-dimensional temporal features to the corresponding predicted value of thermal runaway risk. Exemplarily, the thermal runaway prediction model can be a Long Short-Term Memory (LSTM) network, a Gated Recurrent Unit (GRU), a Temporal Convolutional Network (TCN), or a combination of the above models, but this invention is not limited thereto. When the abnormal feature sequence... After being input into the thermal runaway prediction model, the model calculates the corresponding predicted thermal runaway risk value R through forward propagation, which is expressed as follows: Wherein, f(·) represents the nonlinear mapping function of the thermal runaway prediction model, and the predicted thermal runaway risk value R is a continuous value used to characterize the risk level of thermal runaway of the target battery within the predicted time range. In this embodiment, the predicted thermal runaway risk value can be expressed as the probability of thermal runaway occurrence, with a value range of [0,1].

[0106] In some embodiments, the predicted thermal runaway risk value is compared with a third preset threshold to determine the thermal runaway risk level of the lithium-ion battery based on the third comparison result. Specifically, after obtaining the predicted thermal runaway risk value, the predicted thermal runaway risk value is compared with a pre-set third preset threshold to determine the thermal runaway risk level of the target battery. In this embodiment, the third preset threshold includes at least two thresholds, namely a low-risk threshold T. L and high-risk threshold T H And satisfy: 0≤T L <T H ≤1. Based on the comparison results, determine the thermal runaway risk level of the target battery, specifically including: when R <T L When T is in a low-risk state, the target battery is determined to be in a low-risk state; when T L ≤R <T H When R ≥ T, the target battery is determined to be in a medium-risk warning state; H When the target battery is in a high-risk thermal runaway state, it is determined that the battery is in such a state. Through this method, continuous predicted thermal runaway risks can be transformed into clear risk level results, thus providing the battery management system with intuitive and actionable early warning information about thermal runaway.

[0107] Through the above steps, a unified scale of abnormal feature sequence can be constructed based on the target abnormal operation data. A preset deep learning thermal runaway prediction model can be used to quantitatively predict the future thermal runaway risk of the battery. Combined with threshold judgment, a clear risk level classification can be achieved. Thus, the battery thermal runaway risk can be assessed in real time and accurately, providing a reliable basis for safety management and early warning.

[0108] like Figure 5 As shown, based on the above method embodiments, corresponding apparatus embodiments are provided;

[0109] An embodiment of the present invention provides a schematic diagram of a dynamic prediction system for battery thermal runaway, including: an acquisition module 100, a reconstruction module 200, and a prediction module 300;

[0110] The acquisition module 100 is used to acquire several operational data of the lithium-ion battery;

[0111] The reconstruction module 200 is used to input each of the operating data into a first autoencoder for feature reconstruction to learn the electrothermal coupling change characteristics of each of the operating data that deviate from normal operating conditions, and obtain first abnormal data; input each of the operating data into a second autoencoder for feature reconstruction to learn the abnormal response characteristics of each of the operating data, and obtain second abnormal data; calculate the correlation between each data feature based on the feature distribution relationship between the first abnormal data and the second abnormal data; and filter out thermal runaway early warning feature subset and thermal runaway sensitive feature subset based on the correlation, thereby determining the target operating data for thermal runaway prediction.

[0112] The prediction module 300 is used to predict the thermal runaway of the lithium-ion battery based on the target operating data, and obtain the thermal runaway prediction result of the lithium-ion battery.

[0113] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the battery thermal runaway dynamic prediction method provided by any of the above-described method embodiments of the present invention. For a more detailed workflow and principle of this system, please refer to the relevant descriptions of the above methods.

[0114] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0115] Based on the above embodiments of the dynamic prediction method for battery thermal runaway, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the dynamic prediction method for battery thermal runaway of any embodiment of the present invention.

[0116] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0117] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0118] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0119] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the battery thermal runaway dynamic prediction method described in any of the above-described method embodiments of the present invention.

[0120] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0121] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for dynamic prediction of battery thermal runaway, characterized in that, include: Obtain several operational data points for lithium-ion batteries; Each set of operational data is input into a first autoencoder for feature encoding to obtain first reconstructed data corresponding to each set of operational data, wherein the first autoencoder is trained based on normal battery operating conditions; a first reconstruction error is calculated between each set of operational data and each set of first reconstructed data; the first reconstruction error is compared with a first preset threshold to filter each set of operational data based on the first comparison result and identify first abnormal data; each set of operational data is input into a second autoencoder for feature encoding to obtain second reconstructed data corresponding to each set of operational data, wherein the second autoencoder is trained based on battery thermal runaway conditions; a second reconstruction error is calculated between each set of operational data and each set of second reconstructed data; the second reconstruction error is compared with a second preset threshold to filter each set of operational data based on the second comparison result. Identify the second abnormal data; calculate the intersection of the first abnormal data and the second abnormal data to obtain a target set; determine the data in the first abnormal data excluding the target set as a first set, and determine the data in the second abnormal data excluding the target set as a second set; calculate the feature correlation between the data in the first set and the data in the second set and the data in the target set, respectively, to obtain a first correlation result and a second correlation result; filter the data in the first set according to the first correlation result to obtain a thermal runaway early warning feature subset; filter the data in the second set according to the second correlation result to obtain a thermal runaway sensitive feature subset; determine the target operating data for thermal runaway prediction based on the thermal runaway early warning feature subset, the thermal runaway sensitive feature subset, and the target set; Based on the target operating data, thermal runaway prediction of lithium-ion batteries is performed to obtain the thermal runaway prediction results of lithium-ion batteries.

2. The battery thermal runaway dynamic prediction method as described in claim 1, characterized in that, The process of predicting thermal runaway of lithium-ion batteries based on the target operating data, and obtaining the predicted thermal runaway results for lithium-ion batteries, includes: The target operating data is subjected to feature normalization processing to obtain an abnormal feature sequence, wherein the target operating data is all data in the thermal runaway early warning feature subset, the thermal runaway sensitive feature subset, and the target set; The abnormal feature sequence is input into a preset thermal runaway prediction model for inference and calculation to obtain a predicted value of thermal runaway risk. The predicted thermal runaway risk value is compared with a third preset threshold to determine the thermal runaway risk level of the lithium-ion battery based on the third comparison result.

3. A dynamic prediction system for battery thermal runaway, characterized in that, The system includes: an acquisition module, a reconstruction module, and a prediction module; The acquisition module is used to acquire several operational data of the lithium-ion battery; The reconstruction module is configured to: input each set of operating data into a first autoencoder for feature encoding to obtain first reconstructed data corresponding to each set of operating data, wherein the first autoencoder is trained based on normal battery operating conditions; calculate a first reconstruction error between each set of operating data and each set of first reconstructed data; compare the first reconstruction error with a first preset threshold to filter each set of operating data based on the first comparison result and identify first abnormal data; input each set of operating data into a second autoencoder for feature encoding to obtain second reconstructed data corresponding to each set of operating data, wherein the second autoencoder is trained based on battery thermal runaway conditions; calculate a second reconstruction error between each set of operating data and each set of second reconstructed data; compare the second reconstruction error with a second preset threshold to filter each set of operating data based on the second comparison result. The process involves: filtering to identify second abnormal data; calculating the intersection of the first and second abnormal data to obtain a target set; defining the data in the first abnormal data excluding the target set as a first set, and defining the data in the second abnormal data excluding the target set as a second set; calculating the feature correlation between the data in the first and second sets and the data in the target set to obtain a first correlation result and a second correlation result; filtering the data in the first set based on the first correlation result to obtain a thermal runaway early warning feature subset; filtering the data in the second set based on the second correlation result to obtain a thermal runaway sensitive feature subset; and determining the target operating data for thermal runaway prediction based on the thermal runaway early warning feature subset, the thermal runaway sensitive feature subset, and the target set. The prediction module is used to predict the thermal runaway of the lithium-ion battery based on the target operating data, and obtain the thermal runaway prediction result of the lithium-ion battery.

4. A terminal device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the battery thermal runaway dynamic prediction method as described in any one of claims 1-2.

5. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the dynamic prediction method for battery thermal runaway as described in any one of claims 1-2.

6. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the communication device, the battery thermal runaway dynamic prediction method as described in any one of claims 1 to 2 is implemented.

Images