A time-series predictive method and system for early warning of thermal runaway in ternary lithium batteries

Abstract

This invention relates to a time-series predictive method and system for thermal runaway early warning of ternary lithium batteries, belonging to the field of intelligent early warning. By monitoring battery characteristic parameters and dividing them into time sequences, combined with environmental characteristic parameters, the method predicts the probability of thermal runaway and the abrupt change time of characteristic parameters, and then calculates the unit abrupt change time. Based on this, it optimizes thermal runaway early warning identification resources, and finally performs thermal runaway identification and early warning for each segment of battery characteristic parameters. This invention solves the technical problems of insufficient accuracy and low timeliness in thermal runaway early warning judgment of ternary lithium batteries, effectively improving the accuracy and timeliness of thermal runaway early warning, and providing strong protection for the safe use of ternary lithium batteries.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent early warning, and in particular to a time series prediction-based ternary lithium battery thermal runaway early warning method and system. BACKGROUND

[0002] With the rapid development of electric vehicles, energy storage systems and other fields, ternary lithium batteries have been widely used in these fields due to their high energy density, long cycle life and other advantages. However, the safety hazards of ternary lithium batteries in thermal runaway have become increasingly prominent. Thermal runaway refers to the uncontrolled internal chemical reaction of the battery under overheating conditions, leading to a rapid rise in battery temperature, and even causing a fire or explosion, posing a serious threat to personnel and property safety.

[0003] Currently, the thermal runaway early warning technology for ternary lithium batteries mainly includes simple threshold discrimination method and model analysis method. The simple threshold discrimination method sets threshold values for battery characteristic parameters such as temperature and voltage, and triggers an early warning when the monitored value exceeds the threshold value. However, this method has obvious limitations. On the one hand, the setting of threshold values is often based on experience or limited data, which is difficult to fully reflect the thermal runaway characteristics of the battery under different working conditions; on the other hand, when the battery characteristic parameters approach the threshold value but have not yet exceeded, this method cannot provide early warning of potential thermal runaway risks, resulting in insufficient timeliness of the early warning. The model analysis method is to construct mathematical models or machine learning models to monitor and analyze battery characteristic parameters in real time to predict the probability of thermal runaway. Although this method improves the accuracy of early warning to some extent, it also has some problems, such as the need for a large amount of historical data for model training, and the generalization ability of the model is limited by the diversity and representativeness of the training data. In addition, model analysis often requires high computational resources, affecting the timeliness of the early warning, especially in application scenarios with high real-time requirements, making it difficult to meet the demand.

[0004] In summary, the existing thermal runaway early warning of ternary lithium batteries has the problems of insufficient judgment accuracy and low timeliness, which is difficult to meet the safety requirements in actual applications. SUMMARY

[0005] The present application provides a time series prediction-based ternary lithium battery thermal runaway early warning method and system to solve the technical problems of insufficient judgment accuracy and low timeliness of ternary lithium battery thermal runaway early warning in the prior art.

[0006] The technical solution of the present application to solve the above technical problems is as follows:

[0007] In a first aspect, the present application provides a time sequence prediction ternary lithium battery thermal runaway early warning method, the method comprising: monitoring battery characteristic parameters of a ternary lithium battery to obtain a battery characteristic parameter sequence, and dividing the battery characteristic parameter sequence according to time sequence to obtain a battery characteristic parameter segment sequence; collecting environmental characteristic parameters of the ternary lithium battery in operation to predict a thermal runaway probability of the lithium battery and obtain the thermal runaway probability, and predicting a characteristic parameter mutation of the battery to obtain a predicted mutation time; calculating an actual mutation time of the characteristic parameter mutation of the battery according to the battery characteristic parameter sequence, combining the predicted mutation time to obtain a unit mutation time; and performing thermal runaway early warning identification resource optimization according to the unit mutation time and the thermal runaway probability to obtain optimal thermal runaway early warning identification resources, identifying thermal runaway of each battery characteristic parameter segment to obtain a thermal runaway early warning result.

[0008] In a second aspect, the present application provides a time sequence prediction ternary lithium battery thermal runaway early warning system, the system comprising: a data monitoring module for monitoring battery characteristic parameters of a ternary lithium battery to obtain a battery characteristic parameter sequence, and dividing the battery characteristic parameter sequence according to time sequence to obtain a battery characteristic parameter segment sequence; a parameter collection module for collecting environmental characteristic parameters of the ternary lithium battery in operation to predict a thermal runaway probability of the lithium battery and obtain the thermal runaway probability, and predicting a characteristic parameter mutation of the battery to obtain a predicted mutation time; a time calculation module for calculating an actual mutation time of the characteristic parameter mutation of the battery according to the battery characteristic parameter sequence, combining the predicted mutation time to obtain a unit mutation time; and an identification and early warning module for performing thermal runaway early warning identification resource optimization according to the unit mutation time and the thermal runaway probability to obtain optimal thermal runaway early warning identification resources, identifying thermal runaway of each battery characteristic parameter segment to obtain a thermal runaway early warning result.

[0009] The present application has the following beneficial effects: by monitoring battery characteristic parameters and dividing the battery characteristic parameters according to time sequence, combining environmental characteristic parameters to predict a thermal runaway probability and a characteristic parameter mutation time, then calculating a unit mutation time, and optimizing thermal runaway early warning identification resources according to the unit mutation time, finally identifying thermal runaway of each battery characteristic parameter segment, the accuracy and timeliness of thermal runaway early warning are effectively improved, and a strong guarantee is provided for safe use of ternary lithium batteries. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A flowchart of a time sequence prediction ternary lithium battery thermal runaway early warning method provided by the present application is shown.

[0011] Figure 2 A structure diagram of a time sequence prediction ternary lithium battery thermal runaway early warning system provided by the present application is shown.

[0012] Figure labeling: Data monitoring module 11, parameter acquisition module 12, time calculation module 13, identification and early warning module 14. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0015] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0016] Example 1:

[0017] like Figure 1 As shown, this embodiment of the invention provides a time-series predictive method for early warning of thermal runaway in ternary lithium batteries, the method comprising:

[0018] S10: Monitor the characteristic parameters of the ternary lithium battery to obtain the battery characteristic parameter sequence, and divide it according to the time sequence to obtain the battery characteristic parameter segment sequence.

[0019] Exemplarily, the basic principle of time series prediction is to utilize the patterns and trends in historical time series data to predict the data values at a certain future time point or time period through mathematical models or machine learning algorithms. By collecting, organizing and analyzing the data over a period of time, the periodicity, trend or randomness of the data is identified, and then a prediction model is built based on these rules to infer and predict future data. In the context of ternary lithium battery thermal runaway early warning, time series prediction can be applied to the variation of battery characteristic parameters such as temperature, voltage, etc. with time, so as to predict the risk of thermal runaway that may occur in the future.

[0020] Specifically, in the process of thermal runaway early warning of ternary lithium battery, first, the battery characteristic parameters are continuously monitored, including but not limited to battery temperature, voltage, current and other key indicators, which can directly reflect the working state and health of the battery. Preferably, among them, the temperature monitoring includes cell surface temperature (single point accuracy ±0.5℃), module average temperature (global accuracy ±1℃), covering the range of-30℃~80℃; the voltage monitoring includes single cell voltage (resolution 0.001V), battery pack terminal voltage (accuracy 0.1% FS), monitoring overcharge / overdischarge threshold (such as ternary lithium single overcharge threshold 4.3V, overdischarge threshold 2.5V); the current monitoring includes charging and discharging current (accuracy 0.5% FS), distinguishing between charging positive current and discharging negative current, monitoring short circuit current threshold (>5C for 10ms to determine as abnormal). In addition to the above core parameters, the SOC (state of charge) can be monitored, estimated by ampere-hour integration method combined with Kalman filter, accuracy ±3%; the SOH (state of health) is monitored, based on capacity attenuation (<80% to determine as aging) and internal resistance growth (> initial value 1.5 times to determine as abnormal) comprehensive calculation. The selection of battery characteristic parameters can be adjusted according to actual needs.

[0021] The specific sensor selection and sampling frequency in the preferred embodiment includes: a thin film platinum resistance (Pt100) installed on the contact surface of the cell tab and the shell to monitor the cell temperature, with a sampling frequency set to 10 Hz, and data preprocessing is performed by median filtering (window 3 points) combined with normalization method; a thermocouple (K type) installed on the center heat sink of the module to monitor the module temperature, with a sampling frequency set to 5 Hz, and data preprocessing is performed by sliding average filtering (window 5 points) combined with normalization method; an isolated voltage sensor installed on the positive and negative poles of the cell to monitor the single cell voltage, with a sampling frequency set to 1 Hz, and data preprocessing is performed by Butterworth low-pass filtering (cutoff frequency 0.1 Hz); a Hall current sensor installed on the main circuit of the battery pack to monitor the charge and discharge current, with a sampling frequency set to 10 Hz, and data preprocessing is performed by removing DC offset combined with Z-score standardization method; and a digital temperature and pressure sensor installed on the ventilation port of the battery compartment to monitor the ambient temperature and pressure, with a sampling frequency set to 1 Hz, and data preprocessing is performed by exponential smoothing filtering combined with normalization method. The filter algorithm application includes: 3-point median filtering is used for temperature data to eliminate random noise (such as sensor glitches); 5-point sliding average is used for voltage / current data to suppress high-frequency interference (such as switch power supply ripple). The normalization processing formula is: where μ is the historical mean, σ is the historical standard deviation, and it is suitable for continuous parameters such as voltage, current and temperature.

[0022] Further, by using high-precision sensors and data acquisition, the values of the above parameters can be obtained in real time, and arranged in time sequence, thereby forming a complete battery characteristic parameter sequence. In order to more finely analyze the change trend of the battery state, the continuous parameter sequence needs to be divided according to a certain time interval, for example, cut every second, so that multiple battery characteristic parameter segments are obtained, each parameter segment contains the specific value of the battery characteristic parameter in that time period, and together they constitute a battery characteristic parameter segment sequence. This time sequence segmentation method helps to more accurately capture the subtle changes of the battery state in the subsequent steps, and provides strong data support for thermal runaway early warning.

[0023] S20: Collecting the environmental characteristic parameters of the ternary lithium battery, predicting the thermal runaway probability of the lithium battery, obtaining the thermal runaway probability, and predicting the mutation of the battery characteristic parameters, obtaining the predicted mutation time.

[0024] Preferably, the operating environment characteristic parameters of the ternary lithium battery are collected, which include environmental temperature, pressure, humidity, etc., which have a significant impact on the thermal behavior of the battery. Through high-precision environmental sensors deployed around the battery, these parameter data can be obtained in real time and accurately. Subsequently, using the trained thermal runaway probability prediction model, these environmental characteristic parameters are used as input, and the model outputs the probability value of thermal runaway of the lithium battery under the current environmental conditions based on the correlation between historical data and environmental factors.

[0025] At the same time, in order to predict the abnormal situation of the battery in advance, the battery characteristic parameter mutation prediction also needs to be carried out, which relies on another prediction model, i.e. the battery parameter mutation prediction model. This model also takes environmental characteristic parameters as input, but outputs the time span required for the battery characteristic parameter (such as temperature) to occur a preset mutation amplitude (for example, the temperature change amplitude is greater than or equal to 10 degrees Celsius), i.e. the predicted mutation time. The preset mutation amplitude can be set according to the specific situation. In the preferred embodiment, the temperature mutation can be set as ΔT≥5℃ / min (referring to the temperature sudden rise feature before thermal runaway in the national standard GB / T 39222-2020, the average temperature rise rate is >3℃ / min in the last 5 minutes before thermal runaway, and the safety redundancy value is 5℃ / min); the voltage mutation is set as ΔV≥0.3V / s (corresponding to the initial voltage sudden drop rate of internal short circuit, based on the abuse test data of 18650 battery); the current mutation is set as ΔI≥2C / s (normal current change rate <1C / s under fast charging condition, more than 2C / s is judged as abnormal surge).

[0026] Through these two prediction processes, the system can comprehensively evaluate the safety state of the battery under the current environment, and provide an important basis for subsequent thermal runaway warning.

[0027] S30: According to the battery characteristic parameter sequence, the actual mutation time of the battery characteristic parameter mutation is calculated, and the unit mutation time is obtained by combining the predicted mutation time.

[0028] Further, after obtaining the battery characteristic parameter sequence, in order to accurately grasp the actual time length of the battery characteristic parameter mutation, the sequence needs to be analyzed in depth, by identifying the time when the characteristic parameter (such as temperature, voltage, etc.) in the sequence deviates significantly from the normal range, and calculating the time span experienced from the normal state to the mutation state, so as to obtain the actual mutation time of the battery characteristic parameter mutation.

[0029] However, due to limitations in sensor accuracy, data acquisition frequency, and the complex dynamic changes within the battery, the calculation of the actual abrupt change time may contain certain errors. To improve the accuracy of abrupt change time estimation, the actual abrupt change time is fused with the predicted abrupt change time obtained from environmental characteristic parameters. The fusion method can employ weighted averaging or dynamically adjusting weights based on historical data to fully utilize the complementarity between the prior knowledge of the prediction model and the actual observation data, thereby obtaining a more accurate unit abrupt change time. This time index comprehensively considers the results of prediction and actual observation, providing a more reliable time benchmark for subsequent thermal runaway early warning. For example, if the predicted abrupt change time is 10 seconds, but the actual abrupt change time is measured as 12 seconds due to factors such as sensor delay, a weighted average (e.g., each weight is 0.5) can yield a unit abrupt change time of 11 seconds. This fusion result is closer to the actual situation and helps improve the timeliness and accuracy of thermal runaway early warning.

[0030] The weight determination method for the unit mutation time fusion algorithm is based on dynamically adjusting the weights according to historical prediction errors. The specific formula is as follows: ,in, =|Predicting mutation time t-1 -Actual mutation time t-1 | represents the error at the previous time step. k=0.5 is a hyperparameter that adjusts the rate of change of the weights. α t The value range is (0,1). The larger the error, the higher the weight of the actual mutation time. For example, when error > 5 seconds, α t The value is 0.8, emphasizing trust in real-time data. Therefore, the formula for calculating the unit mutation time is: ,in, To predict mutation time, This represents the actual mutation time.

[0031] S40: Based on the unit mutation time and thermal runaway probability, optimize the thermal runaway early warning identification resources to obtain the optimal thermal runaway early warning discrimination resources, perform thermal runaway identification and early warning for each battery characteristic parameter segment, and obtain thermal runaway early warning results.

[0032] Specifically, after obtaining the unit mutation time and thermal runaway probability, to improve the response speed of the thermal runaway early warning system and ensure its accuracy, the thermal runaway early warning identification resources need to be optimized. Specifically, multiple thermal runaway early warning identification models are first constructed, each capable of independently identifying thermal runaway risks. Then, based on the unit mutation time (i.e., the time required for battery characteristic parameters to change from a normal state to a mutation state) and the thermal runaway probability (i.e., the likelihood of the battery experiencing thermal runaway under current environmental conditions), the number of models participating in the early warning identification is dynamically adjusted. For example, if the unit mutation time is short and the thermal runaway probability is high, the system will increase the number of early warning identification models to process more battery characteristic parameter segments in parallel, thereby shortening the early warning response time; conversely, if the unit mutation time is long and the thermal runaway probability is low, the system will appropriately reduce the number of early warning identification models to save computational resources.

[0033] This optimization process allows us to find an optimal resource configuration for thermal runaway early warning, maximizing the early warning response speed while ensuring accuracy. Finally, this optimally configured early warning identifier is used to perform thermal runaway identification and early warning for each battery characteristic parameter segment. By combining the outputs of each identifier, the final thermal runaway early warning result is obtained, achieving timely and accurate early warning of the thermal runaway risk of ternary lithium batteries.

[0034] In a preferred embodiment, battery characteristic parameters of a ternary lithium battery are monitored to obtain a sequence of battery characteristic parameters. The sequence of battery characteristic parameters is then divided according to time sequence to obtain a sequence of battery characteristic parameter segments. This includes: monitoring battery characteristic parameters of a ternary lithium battery to obtain a sequence of battery characteristic parameters, wherein the battery characteristic parameters include battery temperature; dividing the sequence of battery characteristic parameters according to a preset time window to obtain multiple battery characteristic parameter segments; and arranging these segments to obtain a sequence of battery characteristic parameter segments.

[0035] Optionally, in the initial stage of thermal runaway early warning for ternary lithium batteries, it is necessary to implement comprehensive, multi-time-series monitoring of the battery's characteristic parameters. Among these, battery temperature, as one of the core parameters, is directly related to the internal chemical reaction rate, internal resistance changes, and thermal accumulation effects, making it a key indicator in the monitoring system. Simultaneously, parameters such as voltage, current, and SOC (state of charge) are also included in the monitoring scope to comprehensively reflect the battery's operating status.

[0036] The instantaneous values ​​of battery characteristic parameters can be continuously acquired by a high-precision sensor array and a real-time data acquisition system, and arranged in chronological order to construct a sequence of battery characteristic parameters. This sequence fully records the dynamic process of the evolution of battery characteristic parameters over time.

[0037] To improve the accuracy and efficiency of subsequent analysis, a preset time window is needed to divide the battery characteristic parameter sequence into time segments. For example, if the time window is set to 1 second, all battery characteristic parameters collected within one second will constitute a battery characteristic parameter segment. Multiple consecutive parameter segments are arranged in chronological order to ultimately form a sequence of battery characteristic parameter segments. This division method not only helps to capture short-term fluctuations in battery state but also provides a standardized data input format for subsequent thermal runaway early warning algorithms based on time-series pattern recognition. For example, when the battery temperature shows an abnormal upward trend within several consecutive parameter segments, potential thermal runaway risks can be quickly identified based on sequence analysis.

[0038] In a preferred embodiment, environmental characteristic parameters of the ternary lithium battery are collected to predict the probability of thermal runaway and obtain the thermal runaway probability. The method also includes predicting abrupt changes in battery characteristic parameters to obtain the predicted abrupt change time. This process includes: collecting environmental characteristic parameters within the operating environment of the ternary lithium battery, wherein the environmental characteristic parameters include ambient temperature and pressure; calling a thermal runaway probability predictor and a battery parameter abrupt change predictor; inputting the environmental characteristic parameters into the thermal runaway probability predictor and the battery parameter abrupt change predictor respectively; and predicting the output to obtain the thermal runaway probability and the predicted abrupt change time.

[0039] Specifically, the acquisition and utilization of environmental characteristic parameters are crucial steps in building accurate prediction models. High-precision sensors deployed in the battery's operating environment are used to collect key characteristic parameters such as ambient temperature and pressure in real time. Ambient temperature directly affects the battery's heat dissipation efficiency and internal chemical reaction rate, while ambient pressure can influence its thermal stability by altering the internal gas state. The collected environmental characteristic parameter data, after preprocessing, are input into pre-trained thermal runaway probability predictors and battery parameter mutation predictors, respectively.

[0040] The thermal runaway probability predictor is based on machine learning algorithms, such as neural networks or support vector machines. It integrates parameters like ambient temperature and pressure with historical thermal runaway cases to output the probability of battery thermal runaway under the current environment. For example, if the ambient temperature is consistently near the upper limit of the battery's optimal operating temperature and the pressure fluctuates significantly, the predictor may output a higher probability of thermal runaway. Specifically, taking a neural network as an example, the algorithm architecture consists of a three-layer fully connected neural network: input layer, hidden layer 1, hidden layer 2, and output layer. The input layer includes four features: ambient temperature, ambient pressure, humidity, and charge / discharge rate. Hidden layer 1 has 20 neurons with ReLU activation and a dropout rate of 0.2; hidden layer 2 has 10 neurons with ReLU activation and a dropout rate of 0.1; the output layer has one neuron with Sigmoid activation, outputting a probability value between 0 and 1. During model training, the optimizer is Adam, with an initial learning rate of 10.−3 The loss function decays by 0.9 every 50 rounds. The batch size is 32, and the number of iterations is 200 rounds. The binary cross-entropy loss function is used. The model convergence criterion is that the validation set loss decreases by less than 0.001 for 10 consecutive rounds, and the accuracy is ≥95%.

[0041] Meanwhile, the battery parameter mutation predictor focuses on abnormal changes in battery characteristic parameters, such as temperature and voltage. It predicts the time span required for a preset mutation magnitude (e.g., a temperature surge of ≥15℃) to occur through time series analysis or dynamic threshold detection methods; this is known as the predicted mutation time. For example, if the ambient temperature rises sharply and the pressure increases abnormally, the predictor may output a shorter predicted mutation time, indicating that the battery is about to enter a thermal runaway critical state. Specifically, the algorithm architecture is built using a single-layer LSTM network (processing temporal features) combined with a fully connected layer. The network structure includes: an input layer containing a sequence of environmental feature parameters (time step size of 5, i.e., inputting the environmental data of the most recent 5 seconds); an LSTM layer with 16 memory units, an activation function of tanh, returning the output of the last time step; and an output layer with one neuron, an activation function of linear, outputting the predicted mutation time in seconds. During model training, the optimizer uses RMSprop with a learning rate of 5×10⁻⁶. −4 The batch size is 16, the number of iterations is 150, and the loss function is mean squared error (MSE). The model convergence criterion is that the MSE on the validation set is less than 1.0 (i.e., the prediction time error is less than 1 second) and R² is greater than or equal to 0.9.

[0042] Through the above process, dynamic correlation analysis between environmental characteristic parameters and battery thermal runaway risk was achieved, providing a scientific basis for subsequent thermal runaway early warning and prevention.

[0043] In a preferred embodiment, the training steps of the thermal runaway probability predictor and the battery parameter mutation predictor include: collecting a set of sample environmental feature parameters based on historical ternary lithium battery operating data; collecting the average probability of thermal runaway of ternary lithium batteries under different sample environmental feature parameters and labeling it to obtain a set of sample thermal runaway probabilities; collecting the minimum time for the battery feature parameters of ternary lithium batteries to undergo a preset mutation amplitude under different environmental feature parameters and labeling it to obtain a set of sample predicted mutation times; using machine learning to construct the thermal runaway probability predictor and the battery parameter mutation predictor, wherein the input features of the thermal runaway probability predictor and the battery parameter mutation predictor are environmental feature parameters, and the output features are thermal runaway probability and predicted mutation time, respectively; using the set of sample environmental feature parameters, combined with the set of sample thermal runaway probabilities and the set of sample predicted mutation times, respectively, supervising the training of the thermal runaway probability predictor and the battery parameter mutation predictor until convergence.

[0044] Furthermore, in constructing the thermal runaway probability predictor and the battery parameter mutation predictor, it is necessary to first extract the operating data of ternary lithium batteries within a specific time period from the historical database, covering environmental characteristic parameters such as ambient temperature, pressure, and humidity. These parameters constitute the sample environmental characteristic parameter set. For example, data collected in a high-temperature and high-humidity environment in summer generally has an ambient temperature higher than 35℃ and humidity greater than 70%. These environmental characteristic parameters will serve as the basic input dimensions for subsequent model training.

[0045] Subsequently, by analyzing historical data on cases of battery thermal runaway under different combinations of environmental characteristic parameters, the average probability of thermal runaway for each sample was statistically analyzed and labeled, forming a sample thermal runaway probability set. For example, when the ambient temperature exceeds 45°C and the pressure fluctuates abnormally, the probability of battery thermal runaway may be as high as 30%, and this probability value will be explicitly labeled in the corresponding sample.

[0046] Meanwhile, for cases where battery characteristic parameters (such as temperature) experience a preset abrupt change (e.g., a sudden temperature rise of ≥20℃), the shortest time span for the abrupt change under different environmental characteristic parameters is recorded and labeled as the sample predicted abrupt change time set. For example, under extreme conditions of rapid rise in ambient temperature and sudden increase in pressure, the battery temperature may exceed the safety threshold in just 10 seconds, and this time value will be used as key labeling information.

[0047] Based on the labeled data mentioned above, machine learning frameworks, such as deep neural networks or gradient boosting trees, were used to construct a thermal runaway probability predictor and a battery parameter mutation predictor, respectively. The core design of these two predictors is to use environmental feature parameters as input features and, through the nonlinear mapping relationship within the model, output the two target features: thermal runaway probability and predicted mutation time, respectively.

[0048] During the model training phase, the set of sample environmental feature parameters and the set of sample thermal runaway probabilities are paired and input into the thermal runaway probability predictor. Cross-validation and loss function minimization strategies, such as mean squared error or log loss, are employed to iteratively optimize the model parameters until the model's predictive performance on the validation set converges to a stable state. Similarly, the set of sample environmental feature parameters and the set of sample predicted mutation times are paired and input into the battery parameter mutation predictor. Supervised training is performed using loss functions commonly used in regression tasks (such as Huber loss) to ensure the model's accuracy in predicting mutation times.

[0049] Through this process, an intelligent predictor is finally obtained that can quickly and accurately output the probability of thermal runaway and predict the time of sudden change based on real-time environmental characteristic parameters, providing decision support for the safe operation of ternary lithium batteries.

[0050] In a preferred embodiment, the actual mutation time of the battery characteristic parameter mutation is calculated based on the battery characteristic parameter sequence, and the unit mutation time is obtained by combining the predicted mutation time, including: calculating the minimum time for the battery characteristic parameter to undergo a preset mutation amplitude based on the battery characteristic parameter sequence to obtain the actual mutation time; and calculating the unit mutation time based on the predicted mutation time and the actual mutation time.

[0051] For example, the actual mutation time is quantitatively calculated based on the collected battery characteristic parameter sequence (including continuous records of key parameters such as temperature and voltage changing over time). Specifically, by setting a preset mutation amplitude threshold (e.g., a sudden increase in battery temperature ≥18℃ or a sharp drop in voltage ≥0.5V), the parameter sequence is scanned to identify the moment when the characteristic parameter first exceeds the threshold, and the time span from the start moment to the mutation moment is calculated, i.e., the actual mutation time. For example, if the battery temperature sequence shows that within 100 seconds of continuous monitoring, the temperature steadily rises from 25℃ to 42℃, and then suddenly jumps to 45℃ at the 103rd second, the actual mutation time is recorded as 3 seconds (i.e., the time difference from the 100th second to the 103rd second). Subsequently, the system fuses and analyzes the actual mutation time with the predicted mutation time obtained from the previous prediction using environmental characteristic parameters to generate a unit mutation time. The calculation of the unit mutation time can be based on a weighted average or a dynamic weight adjustment strategy. For example, if the predicted mutation time is 5 seconds, while the actual mutation time is 3 seconds, and based on historical data and current environmental conditions, the predicted and actual values ​​are weighted at 0.4 and 0.6 respectively, then the unit mutation time is calculated as (5 × 0.4 + 3 × 0.6) = 3.8 seconds. This fusion result retains the prediction model's prior judgments on environmental factors while integrating the dynamic feedback from real-time monitoring data, thus more accurately reflecting the true timescale of battery characteristic parameter mutations. For instance, in extreme high-temperature environments, if the prediction model overestimates the mutation time due to abnormal ambient temperature, but actual monitoring data shows that the battery's internal thermal management mechanism effectively delays the mutation, then the unit mutation time will significantly bias towards the actual value, providing a more reliable basis for subsequent early warning decisions.

[0052] In a preferred embodiment, thermal runaway early warning identification resources are optimized based on the unit mutation time and thermal runaway probability to obtain optimal thermal runaway early warning discrimination resources. Thermal runaway identification and early warning are then performed on each battery feature parameter segment, including: constructing a thermal runaway early warning identifier array; randomly selecting and configuring a first number of thermal runaway early warning identifiers; obtaining the time taken for the thermal runaway early warning identifiers under the first number to perform thermal runaway identification and early warning calculations on the battery feature parameter segment, and obtaining a first early warning time; calculating a first thermal runaway early warning score based on the first early warning time, the first number, the unit mutation time, and the thermal runaway probability; continuing to randomly configure a random number of thermal runaway early warning identifiers for quantity optimization to obtain the optimal number with the highest early warning score; randomly selecting the optimal number of thermal runaway early warning identifiers, inputting each battery feature parameter segment, outputting the optimal number of thermal runaway identification results, and filtering the thermal runaway identification results with the highest occurrence ratio to obtain a thermal runaway early warning result, wherein the thermal runaway identification result includes yes or no, and early warning processing is performed when the thermal runaway early warning result is yes.

[0053] Preferably, an array of thermal runaway early warning identifiers with various types or parameter configurations is constructed. These identifiers are built based on algorithms such as machine learning and deep learning, and have the ability to extract risk features from battery characteristic parameter segments and determine whether thermal runaway has occurred. Next, a first number of thermal runaway early warning identifiers are randomly selected to participate in the early warning calculation; for example, 5 identifiers are randomly selected from 20 trained identifiers. After selecting the identifiers, real-time collected battery characteristic parameter segments (such as time-series sequences composed of parameters like temperature and voltage) are input into these identifiers, and the time required for them to complete the thermal runaway identification and early warning calculation and output a result ("yes" or "no") is recorded. This time is used as the first early warning time. For example, if most of the 5 identifiers complete the calculation and output a result within 1 second, the first early warning time is recorded as 1 second.

[0054] Subsequently, based on the first warning time, the first number of identifiers, the unit mutation time (characterizing the urgency of battery characteristic parameter mutations), and the thermal runaway probability (characterizing the likelihood of thermal runaway occurring under the current environment and battery state), a thermal runaway warning scoring function is constructed. For example, the function can comprehensively consider dimensions such as the ratio of warning time to mutation time and the consistency between the warning result and the thermal runaway probability to quantitatively evaluate the warning effectiveness under the current identifier configuration and calculate the first thermal runaway warning score. Specifically, the two-dimensional scoring formula is: Time Score = The higher the value, the faster the response; the ideal value is ≥1; Accurate score = Where N is the number of recognizers, P risk This represents the probability of thermal runaway. Ultimately, the overall score = λ × time score + (1 − λ) × accuracy score, where λ = 0.6, representing the time priority weight, which can be adjusted according to the scenario. T alertThe processing time (in seconds) for the identifier array must satisfy T. alert ≤T unit ×0.8 (with 20% safety time reserved).

[0055] To further optimize early warning resources, a random number of thermal runaway early warning identifiers are configured iteratively, and the aforementioned early warning time recording and scoring calculation process is repeated. For example, in subsequent iterations, 8 and 12 identifiers are selected for testing, and the number of identifiers with the highest early warning score is determined as the optimal number, achieving the best balance between early warning timeliness and resource consumption. After determining the optimal number, this number of thermal runaway early warning identifiers is randomly selected again, and real-time battery characteristic parameter segments are input into these identifiers in parallel to obtain multiple thermal runaway identification results. Then, a majority voting mechanism is used to select the identification result with the highest proportion as the final thermal runaway early warning result. For example, if 9 out of 12 identifiers output "yes," the thermal runaway early warning result is determined to be "yes." When the thermal runaway early warning result is "yes," the system immediately triggers the early warning processing flow, such as starting the battery cooling system, cutting off the power supply, or sending alarm information to maintenance personnel, thereby effectively reducing the risk of thermal runaway accidents and ensuring the safe operation of the battery system.

[0056] In a preferred embodiment, constructing a thermal runaway early warning identifier array includes: collecting a set of sample battery feature parameter segments based on thermal management data of ternary lithium batteries over a historical period; labeling sample thermal runaway identification results based on whether thermal runaway occurs in ternary lithium batteries under different sample battery feature parameter segments, thereby obtaining a set of sample thermal runaway identification results; randomly dividing the set of sample battery feature parameter segments and the set of sample thermal runaway identification results to obtain multiple sets of thermal runaway early warning training data, wherein there is data overlap between every two sets of thermal runaway early warning training data; constructing multiple thermal runaway early warning identifiers using machine learning, wherein the input feature of each thermal runaway early warning identifier is a battery feature parameter segment, and the output feature is a thermal runaway identification result; and supervising the multiple thermal runaway early warning identifiers until convergence using the multiple sets of thermal runaway early warning training data to obtain a thermal runaway early warning identifier array.

[0057] Furthermore, in constructing the thermal runaway early warning and identification array, it is necessary to extract a set of sample battery characteristic parameter segments covering key parameters such as temperature, voltage, and current based on historical thermal management data of ternary lithium batteries. These parameter segments record the battery's operating status under different working conditions in time series form. Subsequently, the samples are labeled according to whether thermal runaway occurred within the corresponding time period of each parameter segment in the historical records. For example, if the battery temperature abnormally spikes and is accompanied by a sudden drop in voltage during the recording period of a certain parameter segment, ultimately leading to a thermal runaway event, then that parameter segment is labeled as "yes" (thermal runaway occurred); conversely, if the battery state is stable and without abnormalities, it is labeled as "no" (thermal runaway did not occur), thus forming a set of sample thermal runaway identification results.

[0058] To improve the model's generalization ability and training efficiency, an overlapping sampling strategy is used to randomly divide the set of sample battery feature parameter segments and the set of sample thermal runaway identification results, generating multiple sets of thermal runaway early warning training data. Each set of data contains partially overlapping parameter segments. For example, multiple subsets are extracted from the complete dataset using the sliding window method, with a certain proportion of duplicate samples retained between adjacent subsets. This ensures that the training data can cover a wide range of operating conditions and enhance the model's ability to identify similar scenarios through data intersection.

[0059] Furthermore, based on the aforementioned training data, multiple thermal runaway early warning identifiers are constructed using machine learning frameworks (such as random forests, gradient boosting trees in ensemble learning, or convolutional neural networks in deep learning). Each identifier takes battery feature parameter segments as input features and outputs thermal runaway identification results ("yes" or "no") through nonlinear mapping.

[0060] During the training phase, multiple sets of training data are used to supervise the training of each recognizer. For example, cross-validation is used to divide the data into training and validation sets. Model parameters (such as the number of trees in a random forest and the weights of a deep learning network) are iteratively optimized to minimize the loss function on the validation set (such as cross-entropy loss) until the model's performance metrics (such as accuracy and recall) on the validation set reach the convergence threshold. For example, when the model's validation set accuracy fluctuates by less than 0.5% over five consecutive iterations, the training is considered to have converged.

[0061] Finally, by integrating multiple independently trained and stable thermal runaway early warning identifiers in parallel, a thermal runaway early warning identifier array is formed. This array can comprehensively utilize the advantages of each identifier and output the final early warning result through voting or weighted fusion mechanism, effectively improving the reliability and robustness of thermal runaway early warning.

[0062] In a preferred embodiment, a first thermal runaway warning score is calculated based on the first warning time, the first quantity, the unit mutation time, and the thermal runaway probability, including: calculating the ratio of the unit mutation time to the first warning time to obtain a first time warning score; calculating the ratio of the first quantity to the thermal runaway probability to obtain a first accurate warning score; and calculating a first thermal runaway warning score based on the first time warning score and the first accurate warning score.

[0063] Specifically, in the thermal runaway early warning effectiveness evaluation system, a first-time early warning score is first constructed based on the unit mutation time (the shortest time required for a battery characteristic parameter to change to a critical state) and the first warning time (the time required for a specific number of thermal runaway early warning identifiers to complete the identification task). This score quantifies the response speed by calculating the ratio of the unit mutation time to the first warning time. For example, if the unit mutation time is 10 seconds and the first warning time is only 1 second, the first-time early warning score is 10. The larger the score, the faster the system responds to potential thermal runaway risks. Conversely, if the first warning time is extended to 5 seconds due to redundant identifier configuration or low algorithm efficiency, the score drops to 2, reflecting a decrease in response timeliness.

[0064] Meanwhile, to assess the accuracy of the early warning, a first accurate early warning score is calculated, which is constructed by the ratio of the first quantity (the number of identifiers currently participating in the early warning) to the thermal runaway probability (characterizing the objective possibility of thermal runaway occurring in the current battery state). For example, when the first quantity is 20 identifiers and the thermal runaway probability is 2%, the first accurate early warning score is 10. This score design is based on probability compensation logic, that is, by increasing the number of identifiers, the risk of false alarms / false alarms is reduced, thereby indirectly improving the accuracy of the early warning. However, it should be noted that if the thermal runaway probability suddenly increases to 10% due to changes in the environment or battery state, the score will be adjusted to 2, indicating that the early warning results need to be evaluated more cautiously in high-risk scenarios.

[0065] Ultimately, the first thermal runaway early warning score is obtained by weighted fusion of the first timely early warning score and the first accurate early warning score. For example, to balance the weights of response speed and accuracy, a normalization coefficient of 0.1 is applied to the first accurate early warning score (because its value is usually large, such as the number of early warning units, which may be tens to hundreds), while the first timely early warning score retains its original weight. The two are then weighted and summed to obtain the comprehensive score. For example, if the first timely early warning score is 8 (unit mutation time 12 seconds, first early warning time 1.5 seconds) and the first accurate early warning score is 50 (first number 50 identifiers, thermal runaway probability 1%), then the first thermal runaway early warning score is 8 + 50 × 0.1 = 13. This scoring system, through dynamic adjustment of weights and normalization strategies, achieves joint optimization of the early warning system's timeliness, accuracy, and resource consumption, providing a quantitative basis for subsequent optimization of the number of early warning identifier arrays.

[0066] The time-series predictive method for thermal runaway early warning of ternary lithium batteries provided in this invention has at least the following technical effects:

[0067] 1. By real-time monitoring of battery characteristic parameters and dividing the parameter segment sequence according to time series, and combining environmental characteristic parameters to construct a two-dimensional input feature system, the joint prediction of thermal runaway probability and parameter mutation time is realized. This breaks through the limitations of traditional single parameter or static threshold early warning. By capturing the dynamic evolution law of battery state through time series parameter sequence, and combining the nonlinear influence of environmental factors on thermal runaway risk, the spatiotemporal resolution and accuracy of the early warning model are significantly improved.

[0068] 2. The concept of unit mutation time is proposed. By calculating the ratio of actual mutation time to predicted mutation time, the response time of the early warning system to battery state mutation is quantitatively evaluated. This index, together with the thermal runaway probability, constructs a dual-objective optimization model to drive the dynamic adjustment of the number of early warning identifier arrays, thereby achieving a dynamic balance between early warning timeliness and computational cost.

[0069] 3. A thermal runaway early warning identification array based on an overlapping sampling strategy was constructed. By randomly dividing the training data and retaining the data intersection, each identifier learned the feature distribution under different operating conditions. Combined with the majority voting mechanism to select high-frequency identification results, the misjudgment caused by data noise or overfitting of a single model was effectively suppressed, and the robustness and interpretability of the early warning results were significantly improved.

[0070] Example 2:

[0071] like Figure 2 As shown, based on the same inventive concept as the time-series prediction method for thermal runaway early warning of ternary lithium batteries provided in Embodiment 1, this embodiment of the invention also provides a time-series prediction system for thermal runaway early warning of ternary lithium batteries, the system comprising:

[0072] The data monitoring module 11 is used to monitor the characteristic parameters of the ternary lithium battery, obtain the battery characteristic parameter sequence, and obtain the battery characteristic parameter segment sequence according to the time sequence.

[0073] The parameter acquisition module 12 is used to acquire environmental characteristic parameters of the ternary lithium battery, predict the probability of thermal runaway of the lithium battery, obtain the probability of thermal runaway, and predict the mutation time of the battery characteristic parameters.

[0074] The time calculation module 13 is used to calculate the actual mutation time of the battery characteristic parameter mutation based on the battery characteristic parameter sequence, and to obtain the unit mutation time by combining the predicted mutation time.

[0075] The identification and early warning module 14 is used to optimize the thermal runaway early warning identification resources based on the unit mutation time and thermal runaway probability, obtain the optimal thermal runaway early warning discrimination resources, perform thermal runaway identification and early warning for each battery characteristic parameter segment, and obtain thermal runaway early warning results.

[0076] Furthermore, the data monitoring module 11 is also used to perform the following steps:

[0077] Battery characteristic parameters of ternary lithium batteries are monitored to obtain a sequence of battery characteristic parameters, including battery temperature. The sequence of battery characteristic parameters is divided into time segments according to a preset time window to obtain multiple battery characteristic parameter segments, which are then arranged to obtain a sequence of battery characteristic parameter segments.

[0078] Furthermore, the parameter acquisition module 12 is also used to perform the following steps:

[0079] Environmental characteristic parameters within the operating environment of a ternary lithium battery are collected, including ambient temperature and pressure. A thermal runaway probability predictor and a battery parameter mutation predictor are invoked. The environmental characteristic parameters are input into the thermal runaway probability predictor and the battery parameter mutation predictor, respectively, and the predicted outputs are used to obtain the thermal runaway probability and the predicted mutation time.

[0080] Furthermore, the parameter acquisition module 12 is also used to perform the following steps:

[0081] Based on historical operating data of ternary lithium batteries, a set of sample environmental characteristic parameters is collected. The average probability of thermal runaway in ternary lithium batteries under different sample environmental characteristic parameters is collected, and a sample thermal runaway probability set is obtained. The minimum time for the battery characteristic parameters of ternary lithium batteries to undergo a preset mutation amplitude under different environmental characteristic parameters is collected, and a sample predicted mutation time set is obtained. Machine learning is used to construct a thermal runaway probability predictor and a battery parameter mutation predictor, wherein the input features of the thermal runaway probability predictor and the battery parameter mutation predictor are environmental characteristic parameters, and the output features are thermal runaway probability and predicted mutation time, respectively. Using the sample environmental characteristic parameter set, combined with the sample thermal runaway probability set and the sample predicted mutation time set, the thermal runaway probability predictor and the battery parameter mutation predictor are subjected to supervised training until convergence.

[0082] Furthermore, the time calculation module 13 is also used to perform the following steps:

[0083] Based on the battery characteristic parameter sequence, calculate the minimum time for the battery characteristic parameters to undergo a preset mutation amplitude, and obtain the actual mutation time; based on the predicted mutation time and the actual mutation time, calculate the unit mutation time.

[0084] Furthermore, the identification and early warning module 14 is also used to perform the following steps:

[0085] Construct a thermal runaway early warning identifier array; randomly select and configure a first number of thermal runaway early warning identifiers; obtain the time taken for the first number of thermal runaway early warning identifiers to perform thermal runaway identification and early warning calculations on battery characteristic parameter segments, and obtain a first early warning time; calculate a first thermal runaway early warning score based on the first early warning time, the first number, the unit mutation time, and the thermal runaway probability; continue to randomly configure a random number of thermal runaway early warning identifiers to optimize the number and obtain the optimal number with the highest early warning score; randomly select the optimal number of thermal runaway early warning identifiers, input each battery characteristic parameter segment, output the optimal number of thermal runaway identification results, filter the thermal runaway identification results with the highest occurrence ratio, and obtain the thermal runaway early warning result, wherein the thermal runaway identification result includes yes or no, and early warning processing is performed when the thermal runaway early warning result is yes.

[0086] Furthermore, the identification and early warning module 14 is also used to perform the following steps:

[0087] Based on historical thermal management data of ternary lithium batteries, a set of sample battery characteristic parameter segments is collected. Depending on whether thermal runaway occurs under different sample battery characteristic parameter segments, the thermal runaway identification results are labeled to obtain a sample thermal runaway identification result set. The sample battery characteristic parameter segment set and the sample thermal runaway identification result set are randomly divided to obtain multiple sets of thermal runaway early warning training data, where there is data overlap between every two sets of thermal runaway early warning training data. Machine learning is used to construct multiple thermal runaway early warning recognizers, where the input features of each thermal runaway early warning recognizer are battery characteristic parameter segments, and the output features are thermal runaway identification results. The multiple sets of thermal runaway early warning training data are used to supervise the training of the multiple thermal runaway early warning recognizers until convergence, resulting in a thermal runaway early warning recognizer array.

[0088] Furthermore, the identification and early warning module 14 is also used to perform the following steps:

[0089] Calculate the ratio of the unit mutation time to the first warning time to obtain a first time warning score; calculate the ratio of the first quantity to the thermal runaway probability to obtain a first accurate warning score; calculate a first thermal runaway warning score based on the first time warning score and the first accurate warning score.

[0090] Through the foregoing detailed description of a time-predictive ternary lithium battery thermal runaway early warning method, those skilled in the art can clearly understand that this embodiment is a time-predictive ternary lithium battery thermal runaway early warning system. As the system disclosed in the embodiment corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section description.

[0091] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for early warning of thermal runaway of ternary lithium batteries with timing prediction, characterized in that, The method comprises: The method comprises: The method comprises: The method comprises: According to the battery characteristic parameter sequence, the actual mutation time of the battery characteristic parameter mutation is calculated, and the unit mutation time is obtained in combination with the predicted mutation time; According to the unit mutation time and the thermal runaway probability, the thermal runaway early warning identification resource optimization is performed to obtain the optimal thermal runaway early warning identification resource, and the thermal runaway identification early warning is performed on each battery characteristic parameter segment to obtain the thermal runaway early warning result, including: An array of thermal runaway early warning identifiers is constructed; A first number of thermal runaway early warning identifiers are randomly selected; The time for the first number of thermal runaway early warning identifiers to perform thermal runaway identification early warning calculation on the battery characteristic parameter segment is obtained to obtain a first early warning time; According to the first early warning time, the first number, the unit mutation time and the thermal runaway probability, a first thermal runaway early warning score is calculated, including: The ratio of the unit mutation time to the first early warning time is calculated to obtain a first time early warning score; The ratio of the first number to the thermal runaway probability is calculated to obtain a first accurate early warning score; According to the first time early warning score and the first accurate early warning score, a first thermal runaway early warning score is calculated; A random number of thermal runaway early warning identifiers are randomly configured to optimize the number and obtain the optimal number with the maximum early warning score; 2. The method of claim 1, wherein the method is a method of predicting thermal runaway of a ternary lithium battery in time, characterized in that, The optimal number of thermal runaway early warning identifiers are randomly selected, and each battery characteristic parameter segment is input to output the optimal number of thermal runaway identification results, the thermal runaway identification result with the highest proportion is screened, and the thermal runaway early warning result is obtained, wherein the thermal runaway identification result includes yes or no, and the early warning processing is performed when the thermal runaway early warning result is yes. The method comprises: The method comprises:

3. The method of claim 1, wherein the method is a method of timing prediction of thermal runaway pre-warning of a ternary lithium battery, characterized in that, The method comprises: The method comprises: The method comprises: The method comprises:

4. The method of claim 3, wherein the method is a method of timing predicting ternary lithium battery thermal runaway early warning, characterized in that, The training steps of the thermal runaway probability predictor and the battery parameter mutation predictor include: According to the ternary lithium battery operation data in the historical time, a sample environment characteristic parameter set is collected; The average probability of thermal runaway of the ternary lithium battery under different sample environment characteristic parameters is collected, and a sample thermal runaway probability set is labeled; The minimum time of the battery characteristic parameter of the ternary lithium battery under different environment characteristic parameters to occur a preset mutation amplitude is collected, and a sample prediction mutation time set is labeled; Machine learning is used to construct a thermal runaway probability predictor and a battery parameter mutation predictor, wherein the input features of the thermal runaway probability predictor and the battery parameter mutation predictor are environment characteristic parameters, and the output features are thermal runaway probability and prediction mutation time respectively; The sample environment characteristic parameter set is used to combine the sample thermal runaway probability set and the sample prediction mutation time set respectively, and the thermal runaway probability predictor and the battery parameter mutation predictor are supervised trained to convergence.

5. The method of claim 1, wherein the method is a method of timing prediction of thermal runaway pre-alarm of ternary lithium batteries, characterized in that, According to the battery characteristic parameter sequence, the actual mutation time of the battery characteristic parameter mutation is calculated, and the unit mutation time is obtained by combining the prediction mutation time, including: According to the battery characteristic parameter sequence, the minimum time of the battery characteristic parameter to occur a preset mutation amplitude is calculated to obtain the actual mutation time; According to the prediction mutation time and the actual mutation time, the unit mutation time is calculated.

6. The method of claim 1, wherein the method is a method of timing predicting ternary lithium battery thermal runaway early warning. A thermal runaway early warning recognizer array is constructed, including: According to the thermal management data of the ternary lithium battery in the historical time, a sample battery characteristic parameter segment set is collected, and according to whether the ternary lithium battery appears thermal runaway under different sample battery characteristic parameter segments, the sample thermal runaway identification result is labeled to obtain a sample thermal runaway identification result set; The sample battery characteristic parameter segment set and the sample thermal runaway identification result set are randomly divided to obtain multiple sets of thermal runaway early warning training data, wherein there is a data intersection between every two sets of thermal runaway early warning training data; Machine learning is used to construct multiple thermal runaway early warning recognizers, wherein the input feature of each thermal runaway early warning recognizer is a battery characteristic parameter segment, and the output feature is a thermal runaway identification result; The multiple sets of thermal runaway early warning training data are used to supervise the training of the multiple thermal runaway early warning recognizers to convergence respectively, and a thermal runaway early warning recognizer array is obtained.

7. A time-series predictive ternary lithium battery thermal runaway pre-warning system, characterized in that, A time sequence prediction ternary lithium battery thermal runaway early warning method for implementing any one of claims 1-6, the system comprises: A data monitoring module for monitoring the battery characteristic parameters of the ternary lithium battery to obtain a battery characteristic parameter sequence, and dividing the battery characteristic parameter sequence into battery characteristic parameter segments according to time sequence; A parameter collection module for collecting the environment characteristic parameters of the ternary lithium battery operation, predicting the thermal runaway probability of the lithium battery to obtain the thermal runaway probability, and predicting the mutation of the battery characteristic parameters to obtain the prediction mutation time; A time calculation module for calculating the actual mutation time of the battery characteristic parameter mutation according to the battery characteristic parameter sequence, and obtaining the unit mutation time by combining the prediction mutation time; An early warning module is configured to perform thermal runaway early warning identification resource optimization based on the unit mutation time and the thermal runaway probability, obtain optimal thermal runaway early warning identification resources, perform thermal runaway early warning identification on each battery characteristic parameter segment, and obtain a thermal runaway early warning result, including: constructing a thermal runaway early warning identifier array; randomly selecting a first number of thermal runaway early warning identifiers; obtaining a first early warning time by calculating the time for the first number of thermal runaway early warning identifiers to perform thermal runaway early warning identification on the battery characteristic parameter segment; calculating a first thermal runaway early warning score based on the first early warning time, the first number, the unit mutation time, and the thermal runaway probability, including: calculating a ratio of the unit mutation time to the first early warning time to obtain a first time early warning score; calculating a ratio of the first number to the thermal runaway probability to obtain a first accurate early warning score; calculating the first thermal runaway early warning score based on the first time early warning score and the first accurate early warning score; continuing to randomly configure a random number of thermal runaway early warning identifiers to optimize the number and obtain an optimal number with the maximum early warning score; randomly selecting the optimal number of thermal runaway early warning identifiers, inputting each battery characteristic parameter segment, and outputting the optimal number of thermal runaway identification results to screen the thermal runaway identification result with the highest proportion to obtain a thermal runaway early warning result, wherein the thermal runaway identification result includes yes or no, and the early warning processing is performed when the thermal runaway early warning result is yes.

Images