A battery thermal runaway identification method and storage medium

Abstract

The present invention discloses a battery thermal runaway identification method and storage medium, which relates to the field of battery safety control technology. The method includes: real-time acquisition of key battery parameters through multiple sensors to obtain multi-dimensional feature data; generating composite features through nonlinear combination of multi-dimensional feature data; updating the importance weight of composite features through real-time calculation of feature change rate based on an adaptive weighting mechanism of data distribution; performing time series prediction of composite features through a recursive neural network, and calculating the risk score of battery thermal runaway according to the time series prediction results of composite features; based on the risk score, dividing the thermal runaway risk into different levels, and triggering corresponding warning signals. Through the combination of multi-dimensional feature data fusion, nonlinear combination analysis and neural network prediction, accurate evaluation and early warning of battery thermal runaway are achieved, which significantly improves the safety, reliability and adaptability of the battery system.

Description

Technical Field

[0001] The present invention relates to the technical field of battery safety control, and in particular to a battery thermal runaway identification method and a storage medium. Background Art

[0002] With the rapid development of electric vehicles, energy storage systems and other applications, battery safety issues have attracted widespread attention. Battery thermal runaway is one of the main causes of battery safety accidents. It is usually caused by a series of irreversible thermal reactions caused by key parameters such as temperature, pressure, and current inside or outside the battery exceeding the safe range, which may lead to battery explosion or combustion. Therefore, how to effectively identify and prevent battery thermal runaway in the early stage has become an important research direction of current battery management technology.

[0003] Existing battery thermal management systems mainly rely on real-time data from a single or a small number of sensors, such as temperature, pressure or current, to monitor the status of the battery. However, these methods have the following shortcomings: delayed response: single sensor data often cannot reflect the complex changes inside the battery in a timely manner, and cannot accurately warn of potential thermal runaway risks at an early stage; lack of multi-dimensional analysis: existing technologies are usually based on single-dimensional data analysis, ignoring the correlation between multiple feature data, which may lead to low accuracy in thermal runaway identification; lack of complex feature analysis capabilities: many existing technologies fail to fully utilize advanced algorithms such as nonlinear analysis and neural networks to mine potential correlations in multi-dimensional feature data, and thus cannot accurately predict thermal runaway risks. Summary of the invention

[0004] Based on the above-mentioned shortcomings of the prior art, an object of the present invention is to provide a battery thermal runaway identification method and a storage medium to solve the above-mentioned technical problems.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for identifying battery thermal runaway, comprising:

[0006] Use multiple sensors to collect key battery parameters in real time and obtain multi-dimensional characteristic data;

[0007] Generate composite features by nonlinearly combining multi-dimensional feature data;

[0008] Adaptive weighting mechanism based on data distribution, which updates the importance weight of composite features by calculating the feature change rate in real time;

[0009] The composite features are predicted in time series by using a recurrent neural network, and the risk score of battery thermal runaway is calculated based on the prediction results of the composite features in time series.

[0010] Based on the risk score, the thermal runaway risk is divided into different levels and the corresponding early warning signals are triggered.

[0011] The present invention is further configured such that the key parameters include internal temperature, surface temperature, internal pressure, internal resistance and terminal voltage; the collected key parameters are standardized, and data noise is eliminated through a filtering algorithm to obtain multi-dimensional feature data.

[0012] The present invention is further configured to generate composite features by nonlinearly combining multi-dimensional feature data, including:

[0013] The nonlinear dependency values ​​between the multi-dimensional feature data pairs are calculated by using a preset nonlinear dependency calculation logic to obtain a nonlinear dependency matrix;

[0014] The parameter pair with the highest nonlinear dependency value is selected through the nonlinear dependency value of each pair of feature data in the nonlinear dependency matrix, wherein the selection logic is: sort all nonlinear dependency values ​​in the nonlinear dependency matrix, and select the feature data pair with the largest dependency value until the feature data is less than two or the remaining nonlinear dependency value is zero.

[0015] The present invention is further configured such that the calculation logic of the nonlinear dependency value between multi-dimensional feature data pairs is: Among them, MI(X i , X j ) is the feature data X i and feature data X j The nonlinear dependence between i , x j ) is the feature data X i and feature data X j The joint probability distribution of the feature data X i and feature data X j At the same time, take the value x i and x j The probability, p(x i ) is the feature data X i The marginal probability distribution of the feature data X i Value x i The probability, p(x j ) is the feature data X j The marginal probability distribution of the feature data X j Value x j probability.

[0016] The present invention is further configured to perform a nonlinear combination of the internal resistance and the terminal voltage, and a nonlinear combination of the internal temperature and the internal pressure according to the parameter selection result;

[0017] The calculation logic of the nonlinear combination of internal resistance and terminal voltage is: Among them, F R_V (t) is the composite characteristic of the nonlinear combination of internal resistance and terminal voltage, R(t) is the internal resistance at time t, V(t) is the terminal voltage at time t, T surface (t) is the surface temperature at time t, T internal (t) is the internal temperature at time t, δ is the weight parameter of the internal resistance term, and ζ is the weight parameter of the voltage term;

[0018] The calculation logic of the nonlinear combination of internal temperature and internal pressure is: Among them, F P_ti (t) is the composite characteristic of the nonlinear combination of internal temperature and internal pressure, P(t) is the internal pressure at time t, T internal (t) is the internal temperature at time t, V(t) is the terminal voltage at time t, γ is the weight parameter of the pressure term, α is the weight parameter of the temperature term, and β is the adjustment parameter of the voltage term.

[0019] The present invention is further configured such that the updating logic of the importance weight of the composite feature is: Among them, w i (t) is the importance weight of the i-th composite feature at time t, ω i and ω j is the initial weight of the i-th and j-th composite features, F i (t) and F j (t) is the value of the i-th and j-th composite features at time t, n is the number of composite features, and η is the adjustment coefficient, which is used to control the influence of the feature change rate on the weight adjustment.

[0020] The present invention is further configured such that, in performing time series prediction on composite features by using a recursive neural network, the training logic of the recursive neural network includes:

[0021] Acquire historical multi-dimensional feature data, generate historical composite features by nonlinearly combining the historical multi-dimensional feature data, and set the historical composite features as a training set and a validation set;

[0022] The historical composite feature is set as the input of the recurrent neural network, and the input data is: F(t) = [F R_V (t),F P_ti (t)];

[0023] Through the forward propagation of the recurrent neural network, the output h(t) at each moment depends on the current input F(t) and the hidden state h(t-1) at the previous moment. In the forward propagation of the recurrent neural network, the forget gate: f t =σ(W f ·[ht-1 ,F(t)]+b f ), f t is the output of the forget gate, W f is the weight matrix of the forget gate, b f is the bias term of the forget gate, σ(·) is the activation function; input gate: i t =σ(W i ·[h t-1 ,F(t)]+b i ),i t is the output of the input gate, W i is the weight matrix of the input gate, b i is the bias term of the input gate; memory unit update: C t =f t ·C t-1 +i t tanh(W C ·[h t-1 ,F(t)]+b C ), C t is the state of the memory unit at time t, C t-1 is the memory cell state at time t-1, W C is the weight matrix of the memory unit, b C is the bias term of the memory unit, tanh(·) is the hyperbolic tangent function; output gate: o t =σ(W o ·[h t-1 ,F(t)]+b o ), o t is the output of the output gate, W o is the weight matrix of the output gate, b o is the bias term of the output gate; hidden state update: h t =o t tanh(C t ), h t is the hidden state at time t;

[0024] The predicted output F′(t+1) is compared with the true value F(t+1) through the validation set, and the prediction loss is calculated, where the loss function is:

[0025] The training data is used to input the composite feature data into the recursive neural network through recursive iteration. At the same time, the more important new weights are iterated until the loss converges to the preset expected value.

[0026] The present invention is further configured such that the calculation logic of the risk score of battery thermal runaway is: Among them, Risk(t+1) is the risk score of battery thermal runaway at time t+1, F ′R_V (t+1) and F ′ P_ti (t+1) is the composite eigenvalue at time t+1, w 1 (t+1) and w 2 (t+1) is the importance weight at time t+1.

[0027] The present invention is further configured to classify the thermal runaway risk into different levels based on the risk score and trigger corresponding warning signals, including:

[0028] When the risk score is greater than a first risk score threshold, marking the thermal runaway risk as high risk, disconnecting the power supply and generating an alarm signal;

[0029] When the risk score is less than or equal to the first risk score threshold and greater than the second risk score threshold, the thermal runaway risk is marked as medium risk, prompting manual inspection and generating an early warning signal;

[0030] When the risk score is less than or equal to a second risk score threshold, the thermal runaway risk is marked as low risk and continuously monitored.

[0031] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor of a computer, the computer is enabled to execute a battery thermal runaway identification method as described in any one of the above items.

[0032] The present invention provides a battery thermal runaway identification method and storage medium. The method collects key battery parameters in real time through multiple sensors to obtain multi-dimensional feature data; generates composite features through nonlinear combination of the multi-dimensional feature data; updates the importance weight of the composite features by calculating the feature change rate in real time based on an adaptive weighting mechanism of data distribution; performs time series prediction on the composite features through a recursive neural network, and calculates the risk score of the battery thermal runaway according to the time series prediction result of the composite features; divides the thermal runaway risk into different levels based on the risk score, and triggers corresponding warning signals, and the beneficial effects produced include:

[0033] 1. Multi-dimensional feature data fusion: The present invention effectively improves the accuracy and reliability of battery status assessment by performing multi-dimensional feature data collection and nonlinear combination analysis on key battery parameters collected by multiple sensors, avoiding the misjudgment or delayed recognition problems caused by reliance on a single data source in the prior art;

[0034] 2. Nonlinear feature combination enhances risk assessment capability: The present invention uses a nonlinear algorithm to comprehensively process multi-dimensional feature data, which can capture complex feature associations and change trends, thereby achieving accurate assessment of battery thermal runaway risks. Compared with existing linear analysis methods, nonlinear combination can more comprehensively reflect the real situation inside the battery;

[0035] 3. Neural network prediction improves the timeliness of early warning: The present invention uses a neural network to perform time series prediction on the risk of thermal runaway of the battery, which can predict the potential risk of thermal runaway in advance, buy more time for taking preventive measures, and improve the safety of the battery system. Compared with the existing early warning system based on a single threshold judgment, the present invention can dynamically track the changing trends of multiple parameters, and the early warning is more accurate;

[0036] 4. Risk level classification and automatic warning mechanism: The present invention calculates the risk factor of battery thermal runaway and dynamically divides different risk levels according to the risk factor. It can automatically trigger a warning signal under high-risk conditions and effectively reduce the probability of battery system runaway by adjusting the risk control strategy in real time, thereby enhancing the safety and reliability of the system.

[0037] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0039] Figure 1 The present invention is a flowchart of a method for identifying thermal runaway of a battery according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0041] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0042] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0043] A battery thermal runaway identification method, such as Figure 1 As shown, including:

[0044] Use multiple sensors to collect key battery parameters in real time and obtain multi-dimensional characteristic data;

[0045] Generate composite features by nonlinearly combining multi-dimensional feature data;

[0046] Adaptive weighting mechanism based on data distribution, which updates the importance weight of composite features by calculating the feature change rate in real time;

[0047] The composite features are predicted in time series by using a recurrent neural network, and the risk score of battery thermal runaway is calculated based on the prediction results of the composite features in time series.

[0048] Based on the risk score, the thermal runaway risk is divided into different levels and the corresponding early warning signals are triggered.

[0049] Specifically, in the battery system, multiple sensors are deployed to monitor key physical and electrochemical parameters. Each sensor will collect specific battery status information in real time. Common sensor types and their monitoring parameters include: temperature sensor: used to collect temperature changes inside and on the surface of the battery; pressure sensor: monitors the gas pressure inside the battery; voltage sensor: used to monitor the changes in the battery terminal voltage in real time; current sensor: records the charging and discharging current of the battery; internal resistance sensor: reflects the battery aging state by monitoring the changes in the battery internal resistance; the above sensors continuously record the key parameters of the battery at each moment, and construct the key parameters into multi-dimensional feature data. The data point of each sensor can be regarded as a dimension, and the data points of multiple sensors are combined to form a multi-dimensional feature; the present invention is further configured that the key parameters include internal temperature, surface temperature, internal pressure, internal resistance and terminal voltage; the key parameters collected are standardized, and the data noise is eliminated by filtering algorithms to obtain multi-dimensional feature data; in actual operation, the data collected by the sensor may contain noise, missing values ​​or abnormal values, so the data needs to be preprocessed. The data preprocessing steps include: Denoising: Use filtering algorithms (such as Kalman filtering or moving average) to smooth data and remove noise; Standardization / normalization: Different sensor data have different dimensions, and the data needs to be standardized so that it is within the same order of magnitude to facilitate subsequent algorithm processing. By collecting battery status information of different dimensions through multiple sensors, it can ensure that the system does not miss any important parameters when monitoring the battery, and provide a comprehensive battery operating status. Since the change in battery status is usually the result of the combined effect of multiple parameters, relying solely on a single parameter may not accurately reflect the true state of the battery. Multi-dimensional feature data can help the system improve recognition accuracy through multi-faceted data fusion.

[0050] The present invention is further configured to generate composite features by nonlinearly combining multi-dimensional feature data, including:

[0051] The nonlinear dependency values ​​between the multi-dimensional feature data pairs are calculated by a preset nonlinear dependency calculation logic to obtain a nonlinear dependency matrix; the present invention is further configured such that the calculation logic of the nonlinear dependency values ​​between the multi-dimensional feature data pairs is: Among them, MI(X i ,X j ) is the feature data X i and feature data X j The nonlinear dependence between i ,x j ) is the feature data X i and feature data X j The joint probability distribution of the feature data X i and feature data Xj At the same time, take the value x i and x j The probability, p(x i ) is the feature data X i The marginal probability distribution of the feature data X i Value x i The probability, p(x j ) is the feature data X j The marginal probability distribution of the feature data X j Value x j probability. Specifically, the above calculation logic uses the preset nonlinear dependency calculation logic to quantify the nonlinear dependencies between multi-dimensional feature data, so as to generate composite features through nonlinear combination. Based on the mutual correlation between features, by calculating their nonlinear dependency values, the most dependent feature pairs are found, and then more composite features are constructed to simplify the number of features; by calculating the nonlinear dependencies between feature data, it can help identify which feature pairs are most important for predicting the battery state. Compared with simple linear correlation analysis, mutual information can capture more complex nonlinear relationships, thereby improving the accuracy of feature selection; through nonlinear dependency analysis, the most dependent feature pairs are selected for combination, and the generated composite features not only contain the information of a single feature, but also reflect the interactive relationship between multiple features. This composite feature is more effective in capturing early signs of battery thermal runaway and can reflect more complex battery state changes.

[0052] Through the nonlinear dependency value of each pair of feature data in the nonlinear dependency matrix, select the parameter pair with the highest nonlinear dependency value, where the selection logic is: sort all nonlinear dependency values ​​in the nonlinear dependency matrix, select the feature data pair with the largest dependency value, until the feature data is less than two or the remaining nonlinear dependency value is zero; specifically, by measuring the nonlinear dependency between different features, identify and select the most relevant feature combination, and then construct a composite feature; by selecting the feature pairs with the strongest nonlinear dependency for combination, the performance of the battery thermal runaway prediction model can be significantly improved. Composite features can not only reflect the state of each feature, but also reflect the interaction between them, which helps to improve the prediction accuracy.

[0053] The present invention is further configured to perform a nonlinear combination of the internal resistance and the terminal voltage, and a nonlinear combination of the internal temperature and the internal pressure according to the parameter selection result;

[0054] The calculation logic of the nonlinear combination of internal resistance and terminal voltage is: Among them, F R_V(t) is the composite characteristic of the nonlinear combination of internal resistance and terminal voltage, R(t) is the internal resistance at time t, V(t) is the terminal voltage at time t, T surface (t) is the surface temperature at time t, T internal (t) is the internal temperature at time t, δ is the weight parameter of the internal resistance term, and ζ is the weight parameter of the voltage term; Specifically, the above calculation logic forms a composite feature F by combining the internal resistance R(t) and the terminal voltage V(t) in a nonlinear manner R_V (t), and through the temperature ratio Adjust the coupling relationship between internal resistance and voltage. The power relationship in the formula is R(t) δ and V(t) ζ It reflects the importance of both to the composite feature. The weight parameters δ and ζ are used to adjust the relative contribution of the two features, and their values ​​range from 0 to 1.

[0055] The calculation logic of the nonlinear combination of internal temperature and internal pressure is: Among them, F P_ti (t) is the composite characteristic of the nonlinear combination of internal temperature and internal pressure, P(t) is the internal pressure at time t, T internal (t) is the internal temperature at time t, V(t) is the terminal voltage at time t, γ is the weight parameter of the pressure term, α is the weight parameter of the temperature term, and β is the adjustment parameter of the voltage term; the above calculation logic is calculated by combining the internal pressure P(t) and the internal temperature T internal (t) Perform nonlinear combination to generate composite feature F P_ti (t), by introducing the exponential term e βV(t) , taking the voltage V(t) as the adjustment factor, the combination of pressure and temperature can reflect the thermal runaway risk of the battery under different voltage conditions. The parameters α, β and γ are used to adjust the importance of the internal temperature, pressure and voltage in the combination, respectively, and the value range is between 0 and 1. By nonlinearly combining key parameters such as internal resistance, terminal voltage, temperature and pressure, the dynamic changes of the battery under complex working conditions can be more comprehensively reflected, especially the early signs when the battery is close to thermal runaway. For example, the rapid changes in temperature difference and pressure are manifestations of intensified chemical reactions inside the battery. The nonlinear combination can capture these changes and issue early warnings; the thermal runaway of the battery is not just a change in a single parameter, but the result of complex interactions between multiple parameters. By introducing nonlinear combinations such as powers, logarithms and exponentials, the complex relationship between these parameters can be captured, and the generated composite features can better describe the internal state of the battery.

[0056] The present invention is further configured such that the updating logic of the importance weight of the composite feature is: Among them, wi (t) is the importance weight of the i-th composite feature at time t, ω i and ω j is the initial weight of the i-th and j-th composite features, F i (t) and F j (t) is the value of the i-th and j-th composite features at time t, n is the number of composite features, and η is the adjustment coefficient, which is used to control the degree of influence of the feature change rate on the weight adjustment; specifically, the core of the above calculation logic is to dynamically adjust the weight of the composite features so that the weights of different composite features at time t can be updated according to their importance. The calculation process combines the change rate of the composite features and the initial weight, and weights and normalizes them through an exponential function to calculate the importance weight of each composite feature at the current moment; the value range of the initial weight is 0 to 1, and the sum of all initial weights is 1; the value range of the adjustment coefficient η is 0.01 to 10, and a smaller η value makes the system insensitive to the change rate, while a larger η value can amplify the impact of the change rate. By adjusting η, the stability and response speed of the system can be controlled. For systems that require fast response, a larger η value can be selected; for systems that require smooth operation, a smaller η value can be selected. By introducing the rate of change and exponential amplification functions, the system can adjust the weights of the composite features according to the dynamic changes of the composite features at each moment, so that the feature weights can be updated in real time over time, enhancing the system's sensitivity to the current state; the rate of change of the composite features can capture the rapid state changes in the system. By assigning higher weights to features with larger rates of change, the system can respond more quickly to drastic changes in the battery state, helping to provide early warning of risks such as thermal runaway.

[0057] The present invention is further configured such that, in performing time series prediction on composite features by using a recursive neural network, the training logic of the recursive neural network includes:

[0058] Acquire historical multi-dimensional feature data, generate historical composite features by nonlinearly combining the historical multi-dimensional feature data, and set the historical composite features as a training set and a validation set;

[0059] The historical composite feature is set as the input of the recurrent neural network, and the input data is: F(t) = [F R_V (t),F P_ti (t)];

[0060] Through the forward propagation of the recurrent neural network, the output h(t) at each moment depends on the current input F(t) and the hidden state h(t-1) at the previous moment. In the forward propagation of the recurrent neural network, the forget gate: f t =σ(W f ·[h t-1 ,F(t)]+bf ), f t is the output of the forget gate, W f is the weight matrix of the forget gate, b f is the bias term of the forget gate, σ(·) is the activation function; input gate: i t =σ(W i ·[h t-1 ,F(t)]+b i ),i t is the output of the input gate, W i is the weight matrix of the input gate, b i is the bias term of the input gate; memory unit update: C t =f t ·C t-1 +i t tanh(W C ·[h t-1 ,F(t)]+b C ), C t is the state of the memory unit at time t, C t-1 is the memory cell state at time t-1, W C is the weight matrix of the memory unit, b C is the bias term of the memory unit, tanh(·) is the hyperbolic tangent function; output gate: o t =σ(W o ·[h t-1 ,F(t)]+b o ), o t is the output of the output gate, W o is the weight matrix of the output gate, b o is the bias term of the output gate; hidden state update: h t =o t tanh(C t ), h t is the hidden state at time t; specifically, in a recurrent neural network, the gating mechanism (forget gate, input gate, output gate) in the network is used to control the flow of information. The forget gate determines which memory information from the previous moment needs to be retained; the input gate determines which feature information of the current input needs to be written into the memory unit; the memory unit updates the memory state at the current moment according to the output of the forget gate and the input gate; the output gate determines which information is output from the memory unit for the current hidden state update;

[0061] The predicted output F′(t+1) is compared with the true value F(t+1) through the validation set, and the prediction loss is calculated, where the loss function is: This loss function measures the gap between the model's predicted value and the true value. By continuously adjusting the model's parameters (including the weight matrix and bias term), the loss function is minimized;

[0062] Using the training data, the composite feature data is input into the recursive neural network through recursive iteration. At the same time, the new weights of importance are iterated until the loss converges to the preset expected value. The core of the above logic is to predict the future battery status based on historical composite feature data of multiple dimensions through the time series prediction function of the recursive neural network. The composite feature data is composed of nonlinear combinations, which can better reflect the complex dynamic behavior inside the battery. These composite features are used as the input of the recursive neural network. After training and verification of the network, the composite feature values ​​at future moments are predicted, and the prediction results are optimized through the loss function, ultimately improving the prediction accuracy; the recursive neural network can use historical data to dynamically predict future states, which is particularly suitable for time series data. Through the analysis of historical composite features, the model can issue early warning signals of battery abnormalities.

[0063] The present invention is further configured such that the calculation logic of the risk score of battery thermal runaway is: Among them, Risk(t+1) is the risk score of battery thermal runaway at time t+1, F ′ R_V (t+1) and F ′ P_ti (t+1) is the composite eigenvalue at time t+1, w 1 (t+1) and w 2 (t+1) is the importance weight at time t+1. Specifically, the above calculation logic constructs a formula based on logistic regression to calculate the risk score Risk(t+1) of battery thermal runaway and predict the probability of battery thermal runaway in the future. The specific formula uses the composite feature F of battery prediction. ′ R_V (t+1) and F ′ P_ti (t+1) and the corresponding weight w 1 (t+1) and w 2(t+1) to calculate the risk score. The logistic regression function in the formula can map the linear combination of composite features to the interval [0,1], and then interpret the result as the probability of battery thermal runaway at a future moment; logistic regression is a regression model commonly used for classification tasks. In this scenario, logistic regression is used to predict the probability of battery thermal runaway. The core idea of ​​logistic regression is to compress the output of the linear model to between 0 and 1 through the sigmoid function, and then interpret the result as a probability; through the logistic regression function, the system can quantify the risk score of battery thermal runaway and map it to a probability interval of 0 to 1. This helps to more intuitively understand the current health status of the battery and facilitates real-time monitoring and management; compared with the use of original features such as temperature, pressure, and internal resistance alone, composite features can better capture the complex interactions within the battery. Through these composite features, the system can improve the accuracy of thermal runaway prediction; the dynamic adjustment mechanism of weights enables the system to flexibly respond to battery states under different working conditions. For example, in the case of large fluctuations in battery temperature, the system can automatically increase the weight of temperature-related composite features to enhance sensitivity to related changes.

[0064] The present invention is further configured to classify the thermal runaway risk into different levels based on the risk score and trigger corresponding warning signals, including:

[0065] When the risk score is greater than a first risk score threshold, marking the thermal runaway risk as high risk, disconnecting the power supply and generating an alarm signal;

[0066] When the risk score is less than or equal to the first risk score threshold and greater than the second risk score threshold, the thermal runaway risk is marked as medium risk, prompting manual inspection and generating an early warning signal;

[0067] When the risk score is less than or equal to the second risk score threshold, the thermal runaway risk is marked as low risk and continuously monitored. Specifically, the above logic divides the thermal runaway risk level based on the risk score, and triggers the corresponding early warning signal according to different risk levels. Specifically, the battery thermal runaway risk score calculated by logistic regression is used as the core indicator, and the risk score will be updated according to the real-time battery status. According to the different risk scores, the system will divide the battery thermal runaway risk into three levels: high risk, medium risk, and low risk, and take corresponding measures such as automatic power off, manual inspection, or continuous monitoring; high risk means that the battery has a greater risk of thermal runaway, triggering emergency measures; medium risk means that there may be a potential risk of runaway, and manual inspection is required; low risk indicates that the battery is in good condition and regular monitoring continues. By calculating the risk score in real time and dividing it into different risk levels, the system can issue an alarm at the early stage of the thermal runaway risk and automatically take protective measures. This greatly improves the system response speed and reduces the risk of battery thermal runaway; different risk levels correspond to different processing methods, automatic power off and alarm at high risk, prompt manual inspection at medium risk, and regular monitoring at low risk. This hierarchical approach can avoid overreaction and effectively prevent safety hazards caused by battery runaway; for low-risk and medium-risk situations, the system can manage through continuous monitoring and early warning signals, reducing reliance on manual intervention and only prompting manual inspection when necessary. This helps to improve the system's automation level and reduce the possibility of human error.

[0068] Another aspect of the present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, a battery thermal runaway identification method as described above is implemented. The computer-readable storage medium may be included in an electronic device, or may exist independently without being assembled into the electronic device.

[0069] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0070] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0071] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0072] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0073] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0074] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0075] In the several embodiments provided in the present application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0076] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0077] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0078] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0079] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for identifying battery thermal runaway, characterized in that: include: Use multiple sensors to collect key battery parameters in real time and obtain multi-dimensional characteristic data; Generate composite features by nonlinearly combining multi-dimensional feature data; Adaptive weighting mechanism based on data distribution, which updates the importance weight of composite features by calculating the feature change rate in real time; The composite features are predicted in time series by using a recurrent neural network, and the risk score of battery thermal runaway is calculated based on the prediction results of the composite features in time series. Based on the risk score, the thermal runaway risk is divided into different levels and the corresponding warning signals are triggered; According to the parameter pair selection result, the internal resistance and the terminal voltage are nonlinearly combined, and the internal temperature and the internal pressure are nonlinearly combined; The calculation logic of the nonlinear combination of internal resistance and terminal voltage is: ,in, It is the composite characteristic of the nonlinear combination of internal resistance and terminal voltage. for The internal resistance of the moment, for The terminal voltage at the moment, for The surface temperature at the moment, for The internal temperature at the time, is the weight parameter of the internal resistance term, is the weight parameter of the voltage term; The calculation logic of the nonlinear combination of internal temperature and internal pressure is: ,in, is a composite feature of the nonlinear combination of internal temperature and internal pressure. for Internal pressure at all times, for The internal temperature at the time, for The terminal voltage at the moment, is the weight parameter of the pressure term, is the weight parameter of the temperature term, is the adjustment parameter of the voltage term; The update logic of the importance weight of the composite feature is: ,in, For the Composite features in The importance weight of the moment, and For the and The initial weights of the composite features, and For the and Composite features in The value of the moment, is the number of composite features, is the adjustment coefficient, which is used to control the influence of the feature change rate on the weight adjustment; In the time series prediction of composite features through recurrent neural networks, the training logic of recurrent neural networks includes: Acquire historical multi-dimensional feature data, generate historical composite features by nonlinearly combining the historical multi-dimensional feature data, and set the historical composite features as a training set and a validation set; The historical composite features are set as the input of the recurrent neural network, and the input data is: ; Through the forward propagation of the recurrent neural network, the output at each moment Depends on current input and the hidden state at the previous moment , where, in the forward propagation of the recurrent neural network, the forget gate: , is the output of the forget gate, is the weight matrix of the forget gate, is the bias term of the forget gate, is the activation function; input gate: , is the output of the input gate, is the weight matrix of the input gate, is the bias term of the input gate; the memory unit is updated: , for The state of the memory unit at the moment, for The state of the memory unit at the moment, is the weight matrix of the memory unit, is the bias term of the memory unit, is the hyperbolic tangent function; output gate: , is the output of the output gate, is the weight matrix of the output gate, is the bias term of the output gate; hidden state update: , for The hidden state of the moment; The prediction output is based on the validation set With the true value Compare and calculate the prediction loss, where the loss function is: ; The training data is used to input the composite feature data into the recursive neural network through recursive iteration. At the same time, the more important new weights are iterated until the loss converges to the preset expected value.

2. A battery thermal runaway identification method according to claim 1, characterized in that: The key parameters include internal temperature, surface temperature, internal pressure, internal resistance and terminal voltage; the collected key parameters are standardized, and data noise is eliminated through a filtering algorithm to obtain multi-dimensional feature data.

3. A battery thermal runaway identification method according to claim 2, characterized in that: Multi-dimensional feature data is nonlinearly combined to generate composite features, including: The nonlinear dependency values ​​between the multi-dimensional feature data pairs are calculated by using a preset nonlinear dependency calculation logic to obtain a nonlinear dependency matrix; The parameter pair with the highest nonlinear dependency value is selected through the nonlinear dependency value of each pair of feature data in the nonlinear dependency matrix, wherein the selection logic is: sort all nonlinear dependency values ​​in the nonlinear dependency matrix, and select the feature data pair with the largest dependency value until the feature data is less than two or the remaining nonlinear dependency value is zero.

4. A battery thermal runaway identification method according to claim 1, characterized in that: The calculation logic of the nonlinear dependency value between multi-dimensional feature data pairs is: ,in, For feature data and feature data The nonlinear dependence between For feature data and feature data The joint probability distribution of and feature data Simultaneous value and The probability of For feature data The marginal probability distribution of Value The probability of For feature data The marginal probability distribution of Value probability.

5. A battery thermal runaway identification method according to claim 1, characterized in that: The calculation logic of the risk score of battery thermal runaway is: ,in, for The risk score of battery thermal runaway at the moment, and for The composite eigenvalue at time, and for The importance weight of the moment.

6. A battery thermal runaway identification method according to claim 5, characterized in that: Based on the risk score, the thermal runaway risk is divided into different levels and the corresponding warning signals are triggered, including: When the risk score is greater than a first risk score threshold, marking the thermal runaway risk as high risk, disconnecting the power supply and generating an alarm signal; When the risk score is less than or equal to the first risk score threshold and greater than the second risk score threshold, the thermal runaway risk is marked as medium risk, prompting manual inspection and generating an early warning signal; When the risk score is less than or equal to a second risk score threshold, the thermal runaway risk is marked as low risk and continuously monitored.

7. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute a battery thermal runaway identification method according to any one of claims 1 to 6.

Images