Battery pack thermal runaway multi-stage early warning method based on VMD-SRM

Abstract

This invention relates to the field of battery pack thermal runaway management, and discloses a multi-level early warning method for battery pack thermal runaway based on VMD-SRM. The method includes constructing a variational model of the original voltage signal based on the VMD algorithm, iteratively updating the modal components, center frequency, and Lagrange multipliers with the sum of modal component bandwidths as the optimization objective, until the convergence condition is met and the final decomposed IMF component is output; evaluating parameters to screen the IMF components; processing the IMF component time series using the SRM algorithm to obtain a state sequence; using standard scores to quantify the deviation of the individual cell state from the overall battery pack state; and providing a thermal runaway safety early warning based on multi-level alarm thresholds. This application can achieve multi-level safety early warning for battery thermal runaway, effectively reducing the interference of inconsistencies between individual cells on diagnostic results and improving the sensitivity of early safety warnings.

Description

Technical Field

[0001] This invention relates to the field of battery pack thermal runaway management, and specifically to a multi-level early warning method for battery pack thermal runaway based on VMD-SRM. Background Technology

[0002] For early thermal runaway warnings based on battery pack electrothermal signals collected by the Battery Management System (BMS) in real-world vehicle environments, effective feature extraction is crucial for ensuring the performance of battery safety warning models. Features are categorized into two types: directly measured features and indirectly processed features. Directly measured features include parameters such as voltage, temperature, and current directly collected by battery sensors. Indirectly processed features, on the other hand, are obtained through parameter identification methods based on directly measured parameters, such as battery internal resistance, state of health, capacity, and self-discharge rate.

[0003] Currently, mainstream battery pack safety warning methods still primarily rely on changes in external characteristic parameters such as voltage and temperature to detect faults, with little in-depth analysis of the underlying fault mechanisms. In fact, voltage signals not only contain information about the normal electrochemical reactions of lithium-ion batteries but also reflect the abnormal state of the battery when a fault occurs. When a battery experiences a short-term voltage anomaly, the signal change amplitude is typically only in the millivolt range, far smaller than the voltage bias caused by inconsistencies between individual cells. This makes traditional fault diagnosis methods based on voltage signal values ​​ineffective in detecting such subtle faults. Summary of the Invention

[0004] This invention aims to provide a multi-level early warning method for battery pack thermal runaway based on VMD-SRM. By combining VMD and SRM algorithms, the voltage signal of individual battery cells is decomposed and the state is identified. Standard scores are used to evaluate the degree of deviation of the state of individual cells, thereby realizing multi-level safety early warning for battery thermal runaway. This effectively reduces the interference of inconsistencies between individual cells on the diagnostic results and improves the sensitivity of early safety warnings.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A multi-level early warning method for battery pack thermal runaway based on VMD-SRM includes:

[0007] S1, Collect the original voltage signal of each battery cell in the battery pack from the BMS;

[0008] S2, based on the VMD algorithm, constructs a variational model of the original voltage signal, takes the sum of the modal component bandwidths as the optimization objective, iteratively updates the modal components, center frequency and Lagrange multipliers until the convergence condition is met, and outputs the final decomposed IMF components;

[0009] S3, Select IMF components for subsequent analysis based on evaluation parameters;

[0010] S4. Based on the time series of the filtered IMF components, the SRM algorithm is used to process the state sequence, which includes the state sequence of individual cells and the overall state sequence of the battery pack.

[0011] S5 uses standard scores to quantify the deviation of the state of a single cell from the state of the overall battery pack.

[0012] S6 provides thermal runaway safety early warning based on multi-level alarm thresholds.

[0013] The principle and advantages of this solution are as follows: In practical applications, battery packs are typically composed of multiple individual cells. Based on the VMD algorithm, the complex original voltage signal is adaptively decomposed into intrinsic mode functions (IMFs) with specific bandwidths, effectively extracting the effective IMF components related to fault characteristics from the original voltage signal of each individual cell. To quantitatively analyze the information content and saliency of each IMF component, IMF components for subsequent analysis are selected based on evaluation parameters. The state sequence is obtained through SRM algorithm processing, which can deeply mine the implicit information related to thermal runaway in the monitoring data, accurately capture the nonlinear thermal characteristics and dynamic failure mechanisms inside the battery, and quantify the deviation of the individual cell state from the overall state of the battery pack using standard scores, thereby judging the battery's health status and potential risks. This application can effectively reduce the interference of inconsistencies between individual cells on diagnostic results, improve the sensitivity of early safety warnings, and enhance the accuracy and reliability of fault detection.

[0014] Preferably, as an improvement, S2 includes:

[0015] S21, Initialize modal components, center frequency, number of iterations, Lagrange multipliers, and construct objective function and constraints;

[0016] S22, introduce a quadratic penalty term and a Lagrange multiplier term to construct an augmented Lagrange function;

[0017] S23 uses the alternating direction multiplier method to solve the augmented Lagrangian function, iteratively updating the modal components, center frequency, and Lagrangian multipliers until the convergence condition is met.

[0018] Technical effect: By continuously optimizing the modal components and center frequency to approach the optimal solution, the voltage signal is ultimately effectively separated in the frequency domain.

[0019] Preferably, as an improvement, the objective function expression includes:

[0020]

[0021] in, For modal components, The order of the decomposition is... The center frequency of each modal component, For partial derivatives, For the Dirac function, For time variables, It is the imaginary unit.

[0022] Technical effect: Facilitates minimizing the sum of bandwidths of all modal components.

[0023] Preferably, as an improvement, the constraints include:

[0024]

[0025] in, For modal components, Let be the decomposition order, and be the constraint condition. This is the original voltage signal.

[0026] Technical effect: It can ensure that the sum of the modal components obtained from the decomposition can accurately reconstruct the original signal.

[0027] Preferably, as an improvement, the augmented Lagrangian function is:

[0028]

[0029]

[0030] in, For modal components, For the center frequency, For Lagrange multipliers, The order of the decomposition is... For partial derivatives, For the Dirac function, For time variables, The imaginary unit, This is the original voltage signal.

[0031] Technical benefit: It facilitates the transformation of constrained problems into unconstrained problems.

[0032] Preferably, as an improvement, the update model for the modal components, center frequency, and Lagrange multipliers is as follows:

[0033]

[0034]

[0035]

[0036]

[0037] in, , , These are used to update the modal components, the center frequency of each mode, and the Lagrange multipliers, respectively. Angular frequency, For the Fourier transform of the original input signal, For the Fourier transform of the i-th modal component, For the Lagrange multipliers, the Fourier transform of the current iteration step. For the Fourier transform of the k-th modal component, For the nth iteration Fourier transform of the Lagrange multipliers, For the update step size of the Lagrange multipliers, For the Fourier transform of the k-th modal component in the (n+1)th iteration, Let be the time-domain representation of the k-th modal component in the (n+1)th iteration. Let be the time-domain representation of the k-th modal component in the nth iteration. This is a preset threshold.

[0038] Technical benefits: Facilitates iterative updates to approximate the optimal solution.

[0039] Preferably, as an improvement, the evaluation parameters include energy percentage and kurtosis value, and the calculation model for the evaluation parameters includes:

[0040]

[0041]

[0042] in, This represents the energy percentage of the i-th IMF component. This represents the value of the i-th IMF component at the n-th sampling point. This represents the value of the original signal at the nth sampling point, where N is the total number of sampling points for the signal. This represents the kurtosis value of the i-th IMF component. Let represent the mean of the i-th IMF component. This represents the standard deviation of the i-th IMF component. It indicates a desire for the expected value.

[0043] Technical effect: By reflecting the contribution of each IMF component to the total energy of the original voltage signal through energy proportion, it can distinguish between low-frequency components with a large proportion and high-frequency components with a small proportion. The kurtosis value measures the sharpness of the signal distribution and can be used to assess the significance of fault information in the signal.

[0044] Preferably, as an improvement, the decomposition order is 2.

[0045] Technical effect: It can decompose the original battery voltage signal into two IMF components. The low-frequency component mainly characterizes the consistency between individual battery cells and the overall aging trend; the high-frequency component can more sensitively capture the dynamic changes inside the battery and effectively reflect the characteristic information related to the fault.

[0046] Preferably, as an improvement, in step S4, a kernel function method is used to define the state function of each individual battery cell voltage, and the overall state function of the battery pack is constructed based on the individual battery cell voltage state function; the state sequence is obtained by processing the IMF component time series based on the state function.

[0047] Technical effect: Facilitates the revelation of complex relationships in the characteristics of individual cell voltages.

[0048] Preferably, as an improvement, the state function includes:

[0049]

[0050]

[0051] in, The state function representing the voltage of a single cell. The overall state function of the battery pack. It is a data matrix containing the voltage characteristics of all individual battery cells. Indicates the first The contribution of each individual cell to the overall state. Represents the kernel function.

[0052] Technical benefits: It can not only assess the overall safety status of the battery pack, but also locate abnormal batteries through the state function values ​​of individual cells. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of the VMD method flow according to an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram of the IMF1 component obtained after the VMD algorithm decomposition of the original voltage signal of the vehicle battery pack in an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of the IMF2 component obtained after the VMD algorithm decomposition of the original voltage signal of the vehicle battery pack in an embodiment of the present invention;

[0057] Figure 5 This is a schematic diagram of the SRM principle according to an embodiment of the present invention;

[0058] Figure 6 This is a schematic diagram of the voltage state of a single battery pack cell according to an embodiment of the present invention;

[0059] Figure 7 This is a schematic diagram of the individual unit voltage state window according to an embodiment of the present invention. Detailed Implementation

[0060] The following detailed description illustrates the specific implementation method:

[0061] The basic implementation examples are as follows: Figure 1 As shown, the multi-level early warning method for battery pack thermal runaway based on VMD-SRM includes:

[0062] S1 collects the original voltage signals of each individual battery cell in the battery pack from the BMS. The battery pack is usually composed of multiple individual battery cells.

[0063] S2 constructs a variational model of the original voltage signal based on the VMD algorithm. Using the sum of the modal component bandwidths as the optimization objective, it iteratively updates the modal components, center frequency, and Lagrange multipliers until the convergence condition is met, outputting the final decomposed IMF components. VMD is a non-recursive signal decomposition technique that adaptively decomposes a complex original voltage signal into a series of intrinsic mode functions (IMFs) with specific bandwidths by constructing and solving a variational problem. The essence of the VMD algorithm lies in solving the optimal solution to a constrained variational problem, i.e., any signal... All can be decomposed into A modal component with finite bandwidth ,in The center frequency of each mode is continuously updated through iterative optimization. And bandwidth, ultimately achieving effective separation of signals in the frequency domain. S2 includes:

[0064] S21, initialize modal components, center frequency, number of iterations, Lagrange multipliers, and construct objective function and constraints.

[0065] In this embodiment, the variational model constructed by VMD aims to minimize the sum of the bandwidths of all modal components; therefore, the objective function expression includes:

[0066]

[0067] in, For modal components, The order of the decomposition is... The center frequency of each modal component, For partial derivatives, For the Dirac function, For time variables, It is the imaginary unit.

[0068] In this embodiment, the constraint condition must satisfy the requirement that the sum of the decomposed modal components must accurately reconstruct the original signal. Therefore, the constraint condition includes:

[0069]

[0070] in, For modal components, Let be the decomposition order, and be the constraint condition. This is the original voltage signal.

[0071] S22, Introducing a quadratic penalty term and Lagrange multipliers, an augmented Lagrange function is constructed to transform the constrained problem into an unconstrained problem. The augmented Lagrange function is:

[0072]

[0073]

[0074] in, For modal components, For the center frequency, For Lagrange multipliers, The order of the decomposition is... For partial derivatives, For the Dirac function, For time variables, The imaginary unit, This is the original voltage signal.

[0075] S23 uses the alternating direction multiplier method to solve for the augmented Lagrangian function, iteratively updating the modal components, center frequency, and Lagrange multipliers until the convergence condition is met. The core objective of the iterative update is to continuously optimize the modal components and center frequency to approximate the optimal solution. The update model for the modal components, center frequency, and Lagrange multipliers (represented in the frequency domain) is as follows:

[0076]

[0077]

[0078]

[0079] in, , , These are used to update the modal components, the center frequency of each mode, and the Lagrange multipliers, respectively. Angular frequency, For the Fourier transform of the original input signal, For the Fourier transform of the i-th modal component, For the Lagrange multipliers, the Fourier transform of the current iteration step. For the Fourier transform of the k-th modal component, For the nth iteration Fourier transform of the Lagrange multipliers, For the update step size of the Lagrange multipliers, Let be the Fourier transform of the k-th modal component in the (n+1)-th iteration (i.e., after the current update). Let be the time-domain representation of the k-th modal component in the (n+1)-th iteration (i.e., after the current update). This is the time-domain representation of the k-th modal component in the nth iteration (i.e., after the current update).

[0080] The convergence criterion is:

[0081]

[0082] in, Let be the time-domain representation of the k-th modal component in the (n+1)th iteration. Let be the time-domain representation of the k-th modal component in the nth iteration. This is a preset threshold. The formula measures the difference between modal components obtained from two adjacent iterations; when the relative change is less than the preset threshold... When convergence is reached, the iteration is terminated, and the final decomposed IMFs are output.

[0083] The VMD algorithm can effectively extract the effective components related to fault characteristics from the raw voltage data of individual cells. The specific implementation process of VMD is as follows: Figure 2 Initialize modal components Center frequency of the mode Lagrange multipliers , Then it enters the iterative calculation process, updating in each iteration. , , and By adaptively adjusting the center frequency of each mode, the system maintains a compact distribution of each mode in the frequency domain and minimizes the spectral overlap between adjacent modes. The iterative process continues until a preset convergence condition is met and a stable IMF component is output.

[0084] In this embodiment, to effectively separate different features in the battery voltage signal, the decomposition order K is set to 2. This setting decomposes the original battery voltage signal into two IMF components: the low-frequency component mainly characterizes the consistency between individual battery cells and the overall aging trend; the high-frequency component can more sensitively capture the dynamic changes inside the battery and effectively reflect fault-related feature information.

[0085] S3. IMF components for subsequent analysis are selected based on evaluation parameters. To quantitatively analyze the information content and saliency of each IMF component, this embodiment further calculates the energy proportion and kurtosis value of each IMF component. The energy proportion reflects the contribution of each component to the total energy of the original signal, distinguishing between low-frequency components with a larger proportion and high-frequency components with a smaller proportion. The kurtosis value measures the sharpness of the signal distribution and can be used to assess the saliency of fault information in the signal.

[0086] The calculation formula is as follows:

[0087]

[0088]

[0089] in, This represents the energy percentage of the i-th IMF component. This represents the value of the i-th IMF component at the n-th sampling point. This represents the value of the original signal at the nth sampling point, where N is the total number of sampling points for the signal. This represents the kurtosis value of the i-th IMF component. Let represent the mean of the i-th IMF component. This represents the standard deviation of the i-th IMF component. It indicates a desire for the expected value.

[0090] by Figure 3 , Figure 4 Taking the battery pack voltage signal decomposition results as an example, the evaluation parameters of each IMF component are calculated as shown in Table 1. IMF1 has a larger energy proportion and a lower kurtosis value, and is identified as a low-frequency component, while IMF2 has a smaller energy proportion and a higher kurtosis value, and is identified as a high-frequency component. Combined with... Figure 3 , Figure 4 The IMF2 exhibits more obvious fluctuations and peaks, consistent with the high kurtosis value. This indicates that the IMF2 can more clearly reflect the rapid changes and potential abnormalities inside the battery. Therefore, the IMF2 is used as the feature component for subsequent fault diagnosis analysis.

[0091] Table 1

[0092]

[0093] S4. Based on the time series of the filtered IMF components, the SRM algorithm is used to process the state sequence, which includes the state sequence of individual cells and the state sequence of the battery pack as a whole.

[0094] State Representation Method (SRM) is a signal-driven, non-parametric analysis technique that constructs a correlation model between monitoring data and the internal state of a system. This allows for the analysis of complex system state characteristics and accurate assessment of health status without relying on precise physical parameter models. Compared to traditional physics-based modeling methods, SRM demonstrates significant advantages in handling the nonlinear characteristics and dynamic state changes of complex systems, overcoming the dependence on high-precision modeling and the limitations of its applicability. In this embodiment, SRM can deeply mine implicit information related to thermal runaway in monitoring data, accurately capturing the nonlinear thermal characteristics and dynamic failure mechanisms within the battery, thereby achieving real-time monitoring and early warning of battery safety performance. The state of a system is its overall response to internal and external factors, specifically depending on the system's structural characteristics and its reaction to the natural environment. The quantitative expression of the system state describes the system's response to external stimuli. If the response meets expectations, the system is considered to be in a normal state; otherwise, it is in an abnormal state. Under normal operating conditions, the system is typically in a steady state, meaning the state remains constant or fluctuates slightly around a certain steady-state value. Under typical operating conditions, the system is generally assumed to be in a stable state. In a general sense, the system state is approximated as a stationary random variable and usually follows a normal distribution.

[0095] The system state is a function of the system's response to environmental stimuli. This assumption has intuitive physical meaning because external stimuli directly affect system behavior. Since the response data is a significant time-varying signal in the time domain, it cannot be used directly and requires feature extraction to transform the signal into the frequency domain or other suitable transform domain. Figure 5 The feature extraction process is demonstrated, in which... Represents the system's eigenvectors. Represents state variables, These are proxy parameters representing the system structure. The system state is represented by variables. The variable is described as a result of characteristic functions extracted from the system's response to various stimuli.

[0096] The successful implementation of the SRM method depends on the system state function. This allows the projection or partial representation of the system state to be revealed through a finite set of measured responses, and all system responses can be considered as the system's state space or characteristic space. It is expressed as follows:

[0097]

[0098] Each system sub-response function All can be viewed as a projection of the system characteristics. In practical applications, if the input of a multi-response system is an excitation signal... We want to obtain the output signal as the test result. , through It is obtained using a time-domain method, the specific method of which is as follows:

[0099]

[0100]

[0101] Sub-response function of the system With excitation signal They are independent of each other. In the process of using SRM, for a target system Able to define a conditional state variable This variable, as a function of the system's state space, is expressed in its frequency domain form as follows:

[0102]

[0103] The essence of SRM is to establish the system response and its system state. The connection between them, if the current system state is normal, means... It meets expectations in terms of safety and reliability. It is initially considered as 1, but over time, the system will gradually deteriorate or malfunction and deviate from its normal state. A value less than 1 indicates a system malfunction.

[0104] Lithium-ion battery packs typically consist of multiple individual cells. The IMF (Integrated Material Function) characteristics of each individual cell reflect its own safety status and directly impact the overall performance of the battery pack. To achieve accurate safety warnings, a system model needs to be constructed that can uniformly characterize the safety status of all individual cells. Assuming the battery pack contains... There are 10 monomers, and the IMF characteristics of each monomer are denoted as . The state representation model construction includes: defining the state function of each individual cell voltage using the kernel function method, constructing the overall state function of the battery pack based on the individual cell voltage state function, and processing the IMF component time series based on the state function to obtain the state sequence.

[0105] Kernel functions map raw features to a high-dimensional space, helping to reveal complex relationships in individual cell voltage characteristics. This embodiment considers the need for a kernel function that can quantify the similarity between individual cell voltage characteristics, thereby capturing subtle deviations in early warning systems. Therefore, a Gaussian kernel function is adopted as shown below:

[0106]

[0107] in, Represents the kernel function.

[0108] The state function of each individual cell voltage is further defined using the kernel function method:

[0109]

[0110] in The weights are obtained through optimization, representing the weights of the first... The contribution of each individual cell to the overall state of the battery pack is illustrated in the following diagram: Figure 6 As shown.

[0111] Based on the individual cell voltage state function, the overall state function of the battery pack can be constructed:

[0112]

[0113] in, It is a data matrix containing the voltage characteristics of all individual cells, and the value of the overall state function is in The state function ranges from 1 to 0, where 1 represents an ideal health state and 0 represents a complete failure. Changes in the state value clearly indicate battery malfunctions. The state function not only assesses the overall safety of the battery pack but also reveals the state function values ​​of individual cells. Locate abnormal batteries. When the health status value of a single battery cell is significantly lower than that of other cells, it can be determined that the cell may have failed or experienced performance degradation.

[0114] S5 uses standard scores to quantify the deviation of the state of a single cell from the state of the overall battery pack, thereby determining the health status and potential risks of the battery.

[0115] The Z-score is a dimensionless parameter defined as the multiple of the standard deviation of a data point from the mean of its dataset. Originally used in statistics to measure the relative position of a data point within a distribution, it quantifies the degree to which a data point deviates from the mean. In this embodiment, by calculating the Z-score of each battery cell's state relative to the states of all cells in the entire battery pack within each time window, it is determined whether that cell deviates from its normal state in the current time window. The Z-score is calculated as follows:

[0116]

[0117] Where Z represents the data point. The Z-score, where x represents the value of that data point. Represents the mean of the dataset. This represents the standard deviation of the dataset.

[0118] S6 provides thermal runaway safety early warning based on multi-level alarm thresholds. For fault alarms in power battery packs, the alarm should not be triggered solely by whether the absolute value of a characteristic parameter at a single time point exceeds a preset threshold. After obtaining features containing fault information, it is necessary to monitor and analyze the changing trends of battery characteristic parameters collected at multiple consecutive time points to make a judgment. An alarm should only be issued when the fault reaches a certain level or duration.

[0119] Therefore, a three-level alarm threshold is set in this embodiment. , and These three thresholds represent: minor battery abnormalities requiring enhanced monitoring; significant battery abnormalities indicating potential risks requiring further analysis and intervention; and severe battery abnormalities with an extremely high risk of thermal runaway, necessitating immediate emergency measures. By statistically analyzing the number of first, second, and third-level alarms within the past 20 windows, the battery fault mode (undervoltage, internal short circuit) can be determined. Figure 7 As shown: The alarm time for cell #2 in car1 was 15:29 on June 4, 2019, which meets the alarm condition of more than 5 level 2 alarms; the alarm time for cell #49 was 18:18 on June 19, 2019, which meets the alarm condition of more than 10 level 1 alarms and more than 3 level 2 alarms.

[0120] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A multi-level early warning method for battery pack thermal runaway based on VMD-SRM, characterized in that, include: S1, Collect the original voltage signal of each battery cell in the battery pack from the BMS; S2, based on the VMD algorithm, constructs a variational model of the original voltage signal, takes the sum of the modal component bandwidths as the optimization objective, iteratively updates the modal components, center frequency and Lagrange multipliers until the convergence condition is met, and outputs the final decomposed IMF components; S3, Select IMF components for subsequent analysis based on evaluation parameters; S4. Based on the time series of the filtered IMF components, the SRM algorithm is used to process the state sequence, which includes the state sequence of individual cells and the overall state sequence of the battery pack. S5 uses standard scores to quantify the deviation of the state of a single cell from the state of the overall battery pack. S6 provides thermal runaway safety early warning based on multi-level alarm thresholds.

2. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 1, characterized in that, S2 includes: S21, Initialize modal components, center frequency, number of iterations, Lagrange multipliers, and construct objective function and constraints; S22, introduce a quadratic penalty term and a Lagrange multiplier term to construct an augmented Lagrange function; S23 uses the alternating direction multiplier method to solve the augmented Lagrangian function, iteratively updating the modal components, center frequency, and Lagrangian multipliers until the convergence condition is met.

3. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 2, characterized in that, The objective function expression includes: in, For modal components, The order of the decomposition is... The center frequency of each modal component, For partial derivatives, For the Dirac function, For time variables, It is the imaginary unit.

4. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 2, characterized in that, The constraints include: in, For modal components, Let be the decomposition order, and be the constraint condition. This is the original voltage signal.

5. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 2, characterized in that, The augmented Lagrange function is: in, For modal components, For the center frequency, For Lagrange multipliers, The order of the decomposition is... For partial derivatives, For the Dirac function, For time variables, The imaginary unit, This is the original voltage signal.

6. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 2, characterized in that, The update model for the modal components, center frequency, and Lagrange multipliers is as follows: in, , , These are used to update the modal components, the center frequency of each mode, and the Lagrange multipliers, respectively. Angular frequency, For the Fourier transform of the original input signal, For the Fourier transform of the i-th modal component, For the Lagrange multipliers, the Fourier transform of the current iteration step. For the Fourier transform of the k-th modal component, For the nth iteration Fourier transform of the Lagrange multipliers, For the update step size of the Lagrange multipliers, For the Fourier transform of the k-th modal component in the (n+1)th iteration, Let be the time-domain representation of the k-th modal component in the (n+1)th iteration. Let be the time-domain representation of the k-th modal component in the nth iteration. This is a preset threshold.

7. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 1, characterized in that, The evaluation parameters include energy percentage and kurtosis value, and the calculation model for the evaluation parameters includes: in, This represents the energy percentage of the i-th IMF component. This represents the value of the i-th IMF component at the n-th sampling point. This represents the value of the original signal at the nth sampling point, where N is the total number of sampling points for the signal. This represents the kurtosis value of the i-th IMF component. Let represent the mean of the i-th IMF component. This represents the standard deviation of the i-th IMF component. It indicates a desire for the expected value.

8. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 3, 4, or 5, characterized in that: The decomposition order is 2.

9. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 1, characterized in that: In step S4, the kernel function method is used to define the state function of each individual cell voltage, and the overall state function of the battery pack is constructed based on the individual cell voltage state function; the state sequence is obtained by processing the IMF component time series based on the state function.

10. The multi-level early warning method for battery pack thermal runaway based on VMD-SRM according to claim 9, characterized in that, The state function includes: in, The state function representing the voltage of a single cell. The overall state function of the battery pack. It is a data matrix containing the voltage characteristics of all individual battery cells. Indicates the first The contribution of each individual cell to the overall state. Represents the kernel function.

Images