Multi-dimensional parameter evaluation system and method based on cell thermal runaway risk detection

Abstract

The invention relates to the technical field of battery management, and discloses a multi-dimensional parameter evaluation system and method based on cell thermal runaway risk detection, a thermal-electric coupling data set is constructed by collecting multi-dimensional parameter data of a cell in different charging and discharging states, and electrochemical impedance spectroscopy data in different aging states are obtained; constructing a refined multi-dimensional electrochemical characteristic space; training a cell health degree evaluation model based on the feature space, extracting a cell health degree index, and establishing a mapping relationship between the cell health degree index and a thermal runaway risk level; in combination with the real-time working condition parameters of the battery cell, a working condition self-adaptive battery cell health degree index is obtained by dynamically adjusting the weight coefficient of each feature in the health degree evaluation model, the thermal runaway risk level is predicted according to the working condition self-adaptive battery cell health degree index, and an early warning mechanism is triggered; according to the invention, accurate, real-time and dynamic evaluation of the thermal runaway risk of the battery cell can be realized, the accuracy and reliability of early warning of the thermal runaway of the battery cell are remarkably improved, and powerful technical support is provided for safety management of the battery.

Description

Technical Field

[0001] The present invention relates to the field of battery management technology, and more specifically, to a multi-dimensional parameter evaluation system and method based on battery cell thermal runaway risk detection. Background Art

[0002] With the widespread application of lithium-ion batteries in electric vehicles, energy storage systems and other fields, their safety issues have become increasingly prominent, especially fires and even explosions caused by battery thermal runaway, which seriously threaten personal safety and property safety. Therefore, developing accurate and reliable thermal runaway risk detection methods and achieving early warning are crucial to ensure the safe operation of lithium-ion batteries. In recent years, domestic and foreign scholars have conducted extensive research on battery thermal runaway risk detection and have made some progress, but there are still some problems that need to be solved.

[0003] Existing research has mainly focused on monitoring and analyzing the external characteristic parameters and internal state parameters of battery thermal runaway. For example, the Chinese patent application with publication number CN117630719A discloses a thermal runaway risk warning system, method, device and medium. The system collects the battery state parameters detected by each battery in the battery pack through the first data acquisition component, and the second data acquisition component collects the thermal runaway element parameters detected by the battery pack. The thermal runaway warning component identifies whether the battery pack has a thermal runaway risk based on the above parameters and issues an alarm when the risk is identified. This method mainly relies on the detection of the external characteristic parameters of the battery (such as voltage, temperature, etc.) and the characteristic gases released when thermal runaway occurs (such as hydrogen, carbon monoxide, etc.). However, the changes in these external characteristic parameters often lag behind the changes in the internal state of the battery cell. When obvious external characteristic changes are detected, the battery may already be in a critical state of thermal runaway, or even thermal runaway has already occurred, making it difficult to achieve effective early warning. The Chinese patent application with publication number CN117103997A proposes a battery thermal runaway risk detection method that takes into account the discharge data of the battery system. The method collects the characteristic data of the battery pack, divides the charge and discharge cycles, extracts the discharge data of a specific cycle, and uses the longitudinal outlier mean algorithm to calculate the thermal runaway risk value of each single battery. The single battery that exceeds the risk threshold is regarded as a suspicious battery, and finally further detection and alarm are performed through the judgment mechanism. Although this method takes into account the discharge data of the battery and quantitatively evaluates the thermal runaway risk, it mainly focuses on the performance of the battery in a specific discharge cycle. It lacks a comprehensive evaluation of the battery under different charge and discharge conditions and different aging conditions, and therefore cannot fully reflect the health status and potential risks of the battery. In addition, neither of the above two methods considers the dynamic characteristics of the battery under actual operating conditions, and the actual operating conditions have an important impact on the health status and thermal runaway risk of the battery.

[0004] The existing technology fails to fully integrate the multi-dimensional parameters that characterize the internal state of the battery cell, especially lacks the analysis and utilization of the electrochemical impedance spectroscopy data of the battery cell in different aging states, resulting in the inability to comprehensively and accurately assess the health status and thermal runaway risk of the battery cell; fails to fully consider the dynamic characteristics of the battery cell under actual operating conditions, lacks real-time monitoring and adaptive adjustment mechanisms for operating parameters, resulting in the inability to accurately assess the thermal runaway risk of the battery cell under actual operating conditions, especially under complex and changeable operating conditions, where this limitation is more obvious. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a multi-dimensional parameter evaluation system and method based on the detection of thermal runaway risk of battery cells. By constructing a refined multi-dimensional electrochemical feature space and combining the working condition adaptive battery cell health index, an accurate, real-time and dynamic evaluation of the thermal runaway risk of battery cells is achieved. This method can not only accurately evaluate the health status of the battery cells, but also effectively predict the thermal runaway risk under different working conditions, significantly improving the accuracy and reliability of the early warning.

[0006] The multi-dimensional parameter evaluation system and method based on thermal runaway risk detection of battery cells proposed in the present invention can be widely used in various application scenarios using lithium-ion batteries as power or energy storage units, such as electric vehicles, buses, energy storage power stations, uninterruptible power supplies, drones, electric bicycles, and household energy storage units. It is particularly suitable for scenarios with extremely high safety requirements, such as large energy storage power stations and electric vehicles.

[0007] To achieve the above object, the present invention provides the following technical solutions: The multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells includes: Synchronously collect multi-dimensional parameter data of battery cells under different charge and discharge states to construct a thermal-electric coupling data set; obtain electrochemical impedance spectroscopy data of battery cells under different aging states, and construct a refined multi-dimensional electrochemical feature space based on the thermal-electric coupling data set and electrochemical impedance spectroscopy data; construct and train a battery cell health assessment model based on the refined multi-dimensional electrochemical feature space, and extract battery cell health indicators; establish a battery cell thermal runaway risk level, and construct a mapping relationship between battery cell health indicators and battery cell thermal runaway risk level; The real-time operating condition parameters of the battery cell are obtained, and the operating condition adaptive battery cell health index is obtained according to the real-time operating condition parameters of the battery cell and the battery cell health assessment model; the predicted thermal runaway risk level is obtained according to the operating condition adaptive battery cell health index and the mapping relationship between the battery cell health index and the thermal runaway risk level; it is determined whether the predicted thermal runaway risk level exceeds a preset threshold, and if so, a warning mechanism is triggered.

[0008] Further, the multi-dimensional parameter data includes multi-point temperature data, load current characteristic data and potential response data; The method of constructing a refined multi-dimensional electrochemical feature space based on the thermal-electric coupling data set and the electrochemical impedance spectroscopy data includes: Based on the thermal-electric coupling data set, the nonlinear frequency response characteristic matrix is ​​obtained; Based on the electrochemical impedance spectroscopy data, the EIS characteristic matrix is ​​obtained; The nonlinear frequency response characteristic matrix and the EIS characteristic matrix are integrated to construct a refined multi-dimensional electrochemical characteristic space.

[0009] Furthermore, the nonlinear frequency response characteristic matrix obtained based on the thermal-electric coupling data set includes: Performing time-frequency analysis on the thermal-electric coupling data set to extract multi-parameter wavelet coefficients; the multi-parameter wavelet coefficients include temperature wavelet coefficients, current wavelet coefficients, and voltage wavelet coefficients; Based on the multi-parameter wavelet coefficients, the nonlinear frequency response characteristic matrix is ​​obtained.

[0010] Further, the extracting multi-parameter wavelet coefficients comprises: Perform time-frequency analysis on multi-point temperature data and extract temperature wavelet coefficients; Perform time-frequency analysis on load current characteristic data and extract current wavelet coefficients; The potential response data is subjected to time-frequency analysis and the voltage wavelet coefficients are extracted.

[0011] Furthermore, the nonlinear frequency response characteristic matrix obtained based on the multi-parameter wavelet coefficients includes: Based on the temperature wavelet coefficient, the energy distribution of multi-point temperature data in different frequency bands is calculated to obtain the temperature frequency domain feature vector; Based on the current wavelet coefficient, the energy distribution of the load current characteristic data in different frequency bands is calculated to obtain the current frequency domain characteristic vector; Based on the voltage wavelet coefficient, the energy distribution of the potential response data in different frequency bands is calculated to obtain the voltage frequency domain feature vector; The temperature frequency domain eigenvector, the current frequency domain eigenvector and the voltage frequency domain eigenvector are concatenated column by column to form a nonlinear frequency response characteristic matrix.

[0012] Furthermore, the EIS characteristic matrix obtained based on the electrochemical impedance spectroscopy data includes: Perform equivalent circuit fitting on electrochemical impedance spectroscopy data to obtain equivalent circuit model parameters; Extract key impedance parameters from equivalent circuit model parameters; The EIS characteristic matrix is ​​constructed from key impedance parameters.

[0013] Furthermore, the fusion of the nonlinear frequency response characteristic matrix and the EIS characteristic matrix to construct a refined multi-dimensional electrochemical characteristic space includes: Perform data standardization on the nonlinear frequency response characteristic matrix and the EIS characteristic matrix; The standardized nonlinear frequency response characteristic matrix and the EIS characteristic matrix are spliced ​​by column to form a multidimensional electrochemical characteristic matrix, wherein each row of the multidimensional electrochemical characteristic matrix corresponds to a battery cell sample, and each column corresponds to an electrochemical characteristic; Perform feature selection on multidimensional electrochemical feature matrix; Feature extraction is performed on the multidimensional electrochemical feature matrix after feature selection to obtain a refined multidimensional electrochemical feature space, wherein the refined multidimensional electrochemical feature space contains n1 electrochemical features.

[0014] Furthermore, the feature selection of the multidimensional electrochemical feature matrix includes: Calculate the importance score of each electrochemical feature in the multidimensional electrochemical feature matrix; Based on the importance score of each electrochemical feature, an electrochemical feature importance matrix is ​​constructed; Based on the electrochemical feature importance matrix, an adaptive threshold method was used to select the optimal feature subset.

[0015] Furthermore, the method of selecting the optimal feature subset based on the electrochemical feature importance matrix by using an adaptive threshold method includes: Design an electrochemical feature selection adaptive threshold function and calculate the electrochemical feature selection adaptive threshold; The electrochemical feature importance matrix is ​​traversed, and the electrochemical features greater than the electrochemical feature selection adaptive threshold are selected into the optimal feature subset.

[0016] Furthermore, the construction and training of the battery cell health assessment model based on the refined multi-dimensional electrochemical feature space includes: The electrochemical features in the refined multi-dimensional electrochemical feature space are used as sample features, and the manually labeled battery cell health indicators are used as sample labels to construct a training sample set; A weighted linear regression model is used to build an initial battery cell health assessment model; Using the training sample set, the initial cell health assessment model is trained to obtain the weight coefficient and bias term b of each electrochemical feature; Based on the weight coefficient and bias term b of each electrochemical feature, the final cell health assessment model is obtained.

[0017] Furthermore, the condition-adaptive battery health index is obtained according to the real-time battery condition parameters and the battery health evaluation model, including: Obtain the operating parameters of the battery cell and construct a mapping relationship between the operating parameters and the weight adjustment coefficient; Calculate the real-time weight adjustment coefficient according to the real-time operating parameters of the battery cell and the mapping relationship between the operating parameters and the weight adjustment coefficient; According to the real-time weight adjustment coefficient, the weight coefficient of each electrochemical feature in the battery cell health assessment model is adjusted to obtain an adjusted weight coefficient; Use the adjusted weight coefficient to calculate the condition-adaptive cell health index under the current condition.

[0018] A multi-dimensional parameter evaluation system based on thermal runaway risk detection of battery cells, which is used to implement the above-mentioned multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells, and the system includes: Feature space construction module: used to synchronously collect multi-dimensional parameter data of battery cells under different charge and discharge states to construct a thermal-electric coupling data set; obtain electrochemical impedance spectroscopy data of battery cells under different aging states, and construct a refined multi-dimensional electrochemical feature space based on the thermal-electric coupling data set and electrochemical impedance spectroscopy data; Mapping relationship establishment module: Based on the refined multi-dimensional electrochemical feature space, build and train the battery cell health assessment model and extract the battery cell health indicators; establish the battery cell thermal runaway risk level and construct the mapping relationship between the battery cell health indicators and the battery cell thermal runaway risk level; Thermal runaway judgment module: obtains the real-time operating parameters of the battery cell, and obtains the operating condition adaptive battery cell health index according to the real-time operating condition parameters of the battery cell and the battery cell health assessment model; obtains the predicted thermal runaway risk level according to the operating condition adaptive battery cell health index and the mapping relationship between the battery cell health index and the thermal runaway risk level; determines whether the predicted thermal runaway risk level exceeds the preset threshold, and if so, triggers the early warning mechanism.

[0019] Compared with the prior art, the present invention has the following beneficial effects: The multidimensional parameter evaluation method based on the detection of thermal runaway risk of battery cells proposed in the present invention innovatively integrates the thermal-electric coupling data and the electrochemical impedance spectroscopy data, constructs a refined multidimensional electrochemical feature space, and comprehensively characterizes the aging state and thermal runaway risk of the battery cells from multiple dimensions. This method uses a weighted linear regression model to construct a health assessment model, and dynamically adjusts the feature weights through operating condition parameters, thereby achieving an accurate assessment of the health status of the battery cells under different operating conditions. In particular, the operating condition adaptive battery cell health index proposed in the present invention can reflect the true health status of the battery cell under the actual operating environment in real time, significantly improving the accuracy and practicality of the health assessment. In addition, the mapping relationship between the battery cell health index and the thermal runaway risk level established by the present invention converts the abstract health index into an intuitive risk level, providing a clear judgment basis for thermal runaway warning. In summary, the multi-dimensional parameter evaluation system and method proposed in the present invention break through the limitations of the traditional single parameter evaluation method, realize the comprehensive, accurate, real-time and dynamic evaluation of the thermal runaway risk of the battery cell, provide strong technical guarantee for battery safety management, effectively reduce the probability of battery thermal runaway accidents, and are of great significance to improving the safety of application fields such as electric vehicles and energy storage systems, and will have a far-reaching impact on promoting the healthy development of the new energy industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 paying creative work.

[0021] Figure 1 It is a principle flow chart of the multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells in the present invention; Figure 2 A flow chart of a method for performing time-frequency analysis on a thermal-electric coupling data set and extracting multi-parameter wavelet coefficients in a multi-dimensional parameter evaluation method based on thermal runaway risk detection of a battery cell according to the present invention; Figure 3 It is a flow chart of a method for obtaining a nonlinear frequency response characteristic matrix based on multi-parameter wavelet coefficients in a multi-dimensional parameter evaluation method based on thermal runaway risk detection of a battery cell according to the present invention; Figure 4 A flow chart of a method for constructing an EIS characteristic matrix in a multidimensional parameter evaluation method based on thermal runaway risk detection of a battery cell according to the present invention; Figure 5A flow chart of a method for integrating a nonlinear frequency response characteristic matrix and an EIS characteristic matrix in a multidimensional parameter evaluation method based on thermal runaway risk detection of a battery cell of the present invention to construct a refined multidimensional electrochemical characteristic space; Figure 6 A flow chart of a method for constructing and training a cell health assessment model in a multi-dimensional parameter assessment method based on cell thermal runaway risk detection of the present invention; Figure 7 A flow chart of a method for obtaining a working condition adaptive battery cell health index in a multi-dimensional parameter evaluation method based on battery cell thermal runaway risk detection of the present invention; Figure 8 It is a flow chart of a method for feature selection of a multidimensional electrochemical feature matrix in a multidimensional parameter evaluation method based on thermal runaway risk detection of a battery cell according to the present invention; Fig. 9 It is a functional module diagram of the multi-dimensional parameter evaluation system based on battery cell thermal runaway risk detection in the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Example 1

[0024] See also Figure 1 As shown, this embodiment provides a multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells, including: Step S1000, synchronously collect multidimensional parameter data of the battery cell under different charge and discharge states, and construct a thermal-electric coupling data set; obtain electrochemical impedance spectroscopy data of the battery cell under different aging states, and construct a refined multidimensional electrochemical feature space based on the thermal-electric coupling data set and the electrochemical impedance spectroscopy data; based on the refined multidimensional electrochemical feature space, construct and train a battery cell health assessment model, and extract battery cell health indicators; establish a battery cell thermal runaway risk level, and construct a mapping relationship between the battery cell health indicator and the battery cell thermal runaway risk level; Furthermore, step S1000 includes: Step S1100, synchronously collecting multi-dimensional parameter data of the battery cell under different charge and discharge states to construct a thermal-electric coupling data set; the multi-dimensional parameter data includes multi-point temperature data, load current characteristic data and potential response data; Specifically, in order to obtain comprehensive temperature distribution information, step S1100 sets a plurality of temperature sensors for collecting temperature distribution data on the surface and inside of the battery cell. These temperature distribution data, as multi-point temperature data, record the spatial distribution characteristics of the battery cell temperature and provide a data basis for the subsequent fusion of temperature parameters. At the same time, the present invention also synchronously collects the load current characteristic data and potential response data of the battery cell. The load current characteristic data reflects the current change law of the battery cell during the charging and discharging process, and reflects the charging and discharging behavior of the battery cell. The potential response data reflects the voltage change law of the battery cell during the charging and discharging process, and reflects the electrochemical characteristics of the battery cell. Through the synchronous collection of load current characteristic data and potential response data, the complex electrochemical process inside the battery cell can be revealed, providing important data support for subsequent feature extraction and health assessment. After the data collection is completed, the present invention preprocesses the collected thermal-electric coupling data set, including removing outliers, performing time synchronization and data alignment and other operations. Data preprocessing can improve data quality, eliminate abnormal interference, and lay a good data foundation for subsequent feature extraction.

[0025] Through the data collection and preprocessing of step S1100, the present invention constructs a comprehensive and reliable thermal-electric coupling data set. This data set comprehensively considers the thermal and electrochemical characteristics of the battery cell, and provides a solid data foundation for subsequent multi-dimensional parameter fusion and health assessment. By synchronously collecting multi-point temperature data, load current characteristic data, and potential response data, the present invention can deeply analyze the physical and chemical processes inside the battery cell and accurately grasp the evolution of the health status of the battery cell. The introduction of data preprocessing further improves the data quality and lays a good foundation for subsequent feature extraction and model construction.

[0026] Step S1200, obtaining a nonlinear frequency response characteristic matrix based on a thermal-electric coupling data set; obtaining electrochemical impedance spectroscopy data of the battery cell under different aging conditions, and obtaining an EIS characteristic matrix based on the electrochemical impedance spectroscopy data; Further, step S1200 includes: Step S1210, obtaining a nonlinear frequency response characteristic matrix based on the thermal-electric coupling data set; Further, step S1210 includes: Step S1211, performing time-frequency analysis on the thermal-electric coupling data set to extract multi-parameter wavelet coefficients; the multi-parameter wavelet coefficients include temperature wavelet coefficients, current wavelet coefficients, and voltage wavelet coefficients; Furthermore, if Figure 2 As shown, step S1211 includes: Step S12111, performing time-frequency analysis on the multi-point temperature data to extract temperature wavelet coefficients; Step S12112, performing time-frequency analysis on the load current characteristic data to extract current wavelet coefficients; Step S12113, perform time-frequency analysis on the potential response data and extract voltage wavelet coefficients.

[0027] Specifically, time-frequency analysis is a powerful signal processing tool that can capture both the time domain and frequency domain characteristics of a signal. Through time-frequency analysis, we can reveal the changing patterns of signals at different time scales and frequency scales and extract the key features of the signal. Wavelet transform is a typical time-frequency analysis method that decomposes the signal at multiple scales to obtain wavelet coefficients in different frequency bands, reflecting the energy distribution characteristics of the signal at different time-frequency scales.

[0028] In sub-step S12111, time-frequency analysis is performed on multi-point temperature data to extract temperature wavelet coefficients. Temperature data reflects the thermal characteristics of the battery cell. By performing wavelet transform on the temperature data, temperature wavelet coefficients of different frequency bands can be obtained. These temperature wavelet coefficients reflect the change pattern of the temperature signal on different time-frequency scales, and reflect the dynamic characteristics of the battery cell temperature. By analyzing the temperature wavelet coefficients, key features such as instantaneous changes and periodic fluctuations of the temperature signal can be captured, providing important information for subsequent temperature feature fusion. In sub-step S12112, time-frequency analysis is performed on the load current feature data to extract current wavelet coefficients. Current data reflects the charging and discharging behavior of the battery cell. By performing wavelet transform on the current data, current wavelet coefficients of different frequency bands can be obtained. These current wavelet coefficients reflect the energy distribution characteristics of the current signal on different time-frequency scales, and reflect the dynamic characteristics of the current. By analyzing the current wavelet coefficients, key features such as instantaneous changes and periodic fluctuations of the current signal can be captured, revealing the dynamic laws of the current during the charging and discharging process of the battery cell, and providing important information for subsequent current feature fusion. In sub-step S12113, the potential response data is subjected to time-frequency analysis to extract the voltage wavelet coefficients. The potential response data reflects the electrochemical characteristics of the battery cell. By performing wavelet transform on the potential response data, the voltage wavelet coefficients of different frequency bands can be obtained. These voltage wavelet coefficients reflect the energy distribution characteristics of the voltage signal on different time-frequency scales, and embody the dynamic characteristics of the voltage. By analyzing the voltage wavelet coefficients, the key features of the instantaneous change and periodic fluctuation of the voltage signal can be captured, revealing the voltage dynamic law during the charging and discharging process of the battery cell, and providing important information for the subsequent voltage feature fusion.

[0029] Through step S1211, the present invention realizes the time-frequency analysis of the thermal-electric coupling data set, and obtains the temperature wavelet coefficient, the current wavelet coefficient and the voltage wavelet coefficient. These wavelet coefficients comprehensively characterize the thermal and electrochemical characteristics of the battery cell from the perspective of time-frequency, and provide rich feature information for the subsequent multi-parameter feature fusion. Through time-frequency analysis, the present invention can deeply explore the dynamic laws contained in the thermal-electric coupling data set, capture the early signs of the evolution of the health status of the battery cell, and achieve a more accurate and comprehensive health assessment.

[0030] Exemplarily, it is assumed that a time-frequency analysis is performed on the thermal-electric coupling data set collected in step S1100. For multi-point temperature data, the db4 wavelet basis function is selected, and the data of each temperature sensor is subjected to a 5-layer wavelet decomposition to obtain temperature wavelet coefficients of different frequency bands. These temperature wavelet coefficients reflect the change pattern of the temperature signal on different time-frequency scales, for example, the low-frequency coefficient reflects the slow change trend of the temperature, and the high-frequency coefficient reflects the rapid fluctuation of the temperature. By analyzing the temperature wavelet coefficients, it is found that the battery cell has obvious high-frequency temperature fluctuations at the end of charging, which may indicate that the battery cell has local overheating and there are safety hazards. For the load current characteristic data, the db4 wavelet basis function is selected, and the current data is subjected to a 5-layer wavelet decomposition to obtain the current wavelet coefficients of different frequency bands. By analyzing the current wavelet coefficients, it is found that when the battery cell is charged and discharged at a large current, the current signal has obvious high-frequency oscillations, which may indicate that a serious polarization phenomenon has occurred inside the battery cell, accelerating the capacity decay. For the potential response data, the sym4 wavelet basis function is selected, and the potential response data is subjected to a 5-layer wavelet decomposition to obtain the voltage wavelet coefficients of different frequency bands. By analyzing the voltage wavelet coefficients, it is found that the voltage signal has obvious low-frequency fluctuations during the charging and discharging process of the battery cell, which may indicate that the internal resistance of the battery cell has increased significantly and the capacity and power characteristics have decreased. Through time-frequency analysis, the degradation law of the health status of the battery cell is revealed from multiple angles, providing an important decision-making basis for subsequent health assessment.

[0031] The application of time-frequency analysis in the present invention greatly expands the analysis dimension of the thermal-electric coupling data set and mines the rich information contained in the data. By extracting the temperature wavelet coefficients, the current wavelet coefficients and the voltage wavelet coefficients, the present invention realizes the dynamic characterization of the thermal and electrochemical characteristics of the battery cell and captures the early signs of the evolution of the health status of the battery cell. This multi-parameter feature extraction method based on time-frequency analysis lays a solid foundation for subsequent feature fusion and health assessment, and helps to improve the accuracy and reliability of health assessment. At the same time, the wavelet coefficients obtained by time-frequency analysis also provide a new perspective for the study of the performance attenuation mechanism of the battery cell, which helps to deeply understand the failure mode and thermal runaway mechanism of the battery cell. In short, the time-frequency analysis method adopted in step S1211 provides a strong technical support for realizing battery cell health assessment and thermal runaway warning based on multi-dimensional parameters.

[0032] Step S1212: obtaining a nonlinear frequency response characteristic matrix based on the multi-parameter wavelet coefficients.

[0033] Furthermore, if Figure 3 As shown, step S1212 includes: Step S12121, based on the temperature wavelet coefficient, calculate the energy distribution of the multi-point temperature data in different frequency bands to obtain the temperature frequency domain feature vector; Step S12122, based on the current wavelet coefficient, calculate the energy distribution of the load current characteristic data in different frequency bands to obtain the current frequency domain characteristic vector; Step S12123, based on the voltage wavelet coefficients, calculating the energy distribution of the potential response data in different frequency bands to obtain the voltage frequency domain feature vector; Step S12124, concatenate the temperature frequency domain characteristic vector, the current frequency domain characteristic vector and the voltage frequency domain characteristic vector by column to form a nonlinear frequency response characteristic matrix.

[0034] Specifically, the nonlinear frequency response characteristic matrix is ​​a characteristic representation method that comprehensively characterizes the nonlinear dynamic behavior of the battery cell. It integrates the frequency domain characteristics of the three parameters of temperature, current, and voltage, and comprehensively reflects the thermal and electrochemical characteristics of the battery cell. In sub-step S12121, based on the temperature wavelet coefficient, the energy distribution of multi-point temperature data in different frequency bands is calculated to obtain the temperature frequency domain characteristic vector. The temperature frequency domain characteristic vector includes the temperature fundamental frequency energy, the temperature second harmonic energy, the temperature third harmonic energy, etc., which reflects the nonlinear frequency response characteristics of the temperature signal. Among them, the temperature fundamental frequency energy represents the energy of the temperature signal at the fundamental frequency (that is, the lowest frequency component of the signal), reflecting the overall change trend of the temperature; the temperature second harmonic energy represents the energy of the temperature signal at twice the fundamental frequency, reflecting the second-order nonlinear effect of the temperature; the temperature third harmonic energy represents the energy of the temperature signal at three times the fundamental frequency, reflecting the third-order nonlinear effect of the temperature. By calculating the temperature frequency domain characteristic vector, the nonlinear frequency response characteristics of the temperature signal can be quantitatively evaluated to reveal the high-order dynamic laws of temperature changes. In sub-step S12122, based on the current wavelet coefficients, the energy distribution of the load current characteristic data in different frequency bands is calculated to obtain the current frequency domain characteristic vector. The current frequency domain characteristic vector includes the current fundamental frequency energy, the current second harmonic energy, the current third harmonic energy, etc., which reflects the nonlinear frequency response characteristics of the current signal. Among them, the current fundamental frequency energy represents the energy of the current signal at the fundamental frequency, reflecting the overall change trend of the current; the current second harmonic energy represents the energy of the current signal at twice the fundamental frequency, reflecting the second-order nonlinear effect of the current; the current third harmonic energy represents the energy of the current signal at three times the fundamental frequency, reflecting the third-order nonlinear effect of the current. By calculating the current frequency domain characteristic vector, the nonlinear frequency response characteristics of the current signal can be quantitatively evaluated to reveal the high-order dynamic laws of current changes. In sub-step S12123, based on the voltage wavelet coefficients, the energy distribution of the potential response data in different frequency bands is calculated to obtain the voltage frequency domain characteristic vector. The voltage frequency domain characteristic vector includes the voltage fundamental frequency energy, the voltage second harmonic energy, the voltage third harmonic energy, etc., reflecting the nonlinear frequency response characteristics of the voltage signal. Among them, the voltage fundamental frequency energy represents the energy of the voltage signal at the fundamental frequency, reflecting the overall change trend of the voltage; the voltage second harmonic energy represents the energy of the voltage signal at twice the fundamental frequency, reflecting the second-order nonlinear effect of the voltage; the voltage third harmonic energy represents the energy of the voltage signal at three times the fundamental frequency, reflecting the third-order nonlinear effect of the voltage. By calculating the voltage frequency domain eigenvector, the nonlinear frequency response characteristics of the voltage signal can be quantitatively evaluated to reveal the high-order dynamic laws of voltage changes.

[0035] In sub-step S12124, the temperature frequency domain eigenvector, the current frequency domain eigenvector and the voltage frequency domain eigenvector are spliced ​​by column to form a nonlinear frequency response characteristic matrix. The nonlinear frequency response characteristic matrix integrates the frequency domain characteristics of the three parameters of temperature, current and voltage, and comprehensively characterizes the nonlinear dynamic behavior of the battery cell. By constructing a nonlinear frequency response characteristic matrix, the frequency domain characteristics of different physical quantities can be integrated into the same mathematical framework, which is convenient for subsequent feature fusion and analysis. The nonlinear frequency response characteristic matrix not only reflects the frequency domain characteristics of a single parameter, but also reveals the frequency domain coupling effect between different parameters, providing a new perspective for in-depth understanding of the thermal-electric coupling behavior of the battery cell. Through step S1212 and its sub-steps, the present invention constructs a nonlinear frequency response characteristic matrix, which realizes a comprehensive characterization of the nonlinear dynamic behavior of the battery cell. The nonlinear frequency response characteristic matrix integrates the frequency domain characteristics of the three parameters of temperature, current and voltage, reveals the high-order dynamic laws of the thermal and electrochemical characteristics of the battery cell, and lays the foundation for the subsequent multi-dimensional electrochemical characteristic fusion. Compared with traditional time domain features, the nonlinear frequency response feature matrix mines the inherent laws of the battery cell state evolution from a frequency domain perspective, providing a more comprehensive and fine-grained feature representation. By analyzing the nonlinear frequency response feature matrix, early signs of battery cell health degradation can be captured, achieving more accurate and timely health status assessment.

[0036] Exemplarily, it is assumed that data collection is performed on a cell, and time series data of temperature, current, and voltage are obtained. Through step S1211, these data are subjected to wavelet transformation to obtain temperature wavelet coefficients, current wavelet coefficients, and voltage wavelet coefficients. In sub-step S12121, based on the temperature wavelet coefficients, the energy distribution of the temperature signal in different frequency bands is calculated, and the temperature frequency domain eigenvector is obtained. For example, the energy of the temperature signal on the fundamental frequency accounts for 80%, the energy on the second harmonic accounts for 15%, and the energy on the third harmonic accounts for 5%. This shows that the temperature signal is mainly composed of fundamental frequency components, but there are also certain nonlinear effects. In sub-steps S12122 and S12123, similar analysis is performed on the current signal and the voltage signal to obtain the current frequency domain eigenvector and the voltage frequency domain eigenvector. Finally, in sub-step S12124, the three frequency domain eigenvectors are spliced ​​by column to form a nonlinear frequency response characteristic matrix. By analyzing the nonlinear frequency response characteristic matrix, it is found that the cell has a significant increase in the energy of the second harmonic of temperature at the end of charging, and the energy of the third harmonic of current also increases. This phenomenon indicates that serious electrochemical polarization may have occurred inside the battery cell at high SOC, and the risk of thermal runaway has increased sharply. Based on this analysis result, the charging strategy can be adjusted in time to prevent the battery cell from entering a high-risk state.

[0037] In short, the nonlinear frequency response characteristic matrix constructed in step S1212 provides new ideas and methods for the assessment of the health status of battery cells. By integrating the frequency domain characteristics of the three parameters of temperature, current, and voltage, the nonlinear frequency response characteristic matrix realizes a comprehensive characterization of the nonlinear dynamic behavior of the battery cell and reveals the inherent laws of the evolution of the battery cell state. This characteristic matrix not only enhances the accuracy and reliability of the health status assessment, but also lays a solid data foundation for the thermal runaway warning of the battery cell. In practical applications, the nonlinear frequency response characteristic matrix can be combined with other electrochemical characteristics to build a more complete battery cell health management system to ensure the safe operation of the battery system.

[0038] Step S1220, obtaining electrochemical impedance spectroscopy data of the battery cell under different aging conditions; Step S1230, extracting key impedance parameters from electrochemical impedance spectroscopy data and constructing an EIS characteristic matrix; Furthermore, if Figure 4 As shown, step S1230 includes: Step S1231, performing equivalent circuit fitting on the electrochemical impedance spectroscopy data to obtain equivalent circuit model parameters; Step S1232, extracting key impedance parameters from equivalent circuit model parameters; Step S1233, constructing an EIS characteristic matrix based on key impedance parameters.

[0039] Specifically, electrochemical impedance spectroscopy is a powerful electrochemical testing technology that can quantitatively evaluate the electrochemical process inside the battery cell and reveal the evolution law of the health status of the battery cell. In the EIS test, a small sinusoidal AC excitation signal is applied to the battery cell, and the impedance values ​​at different frequencies are obtained by measuring the response signal of the battery cell, and then the electrochemical impedance spectrum of the battery cell is drawn. The electrochemical impedance spectrum is usually composed of a high-frequency region, a medium-frequency region, and a low-frequency region, reflecting the characteristics of the ohmic impedance, charge transfer impedance, and diffusion impedance inside the battery cell. In order to comprehensively evaluate the aging state of the battery cell, the present invention collects EIS data at different life cycle stages of the battery cell. Specifically, at key nodes such as the initial state of the battery cell, after 200 cycles, and after 500 cycles, the battery cell is subjected to EIS testing to obtain electrochemical impedance spectroscopy data under different aging states. By tracking the evolution trend of the EIS data, the aging behaviors of the battery cell such as capacity attenuation, internal resistance increase, and power reduction can be quantitatively evaluated, providing important data support for subsequent health status assessment. When collecting EIS data, the test conditions need to be strictly controlled to ensure the consistency and reliability of the data. Factors such as test temperature, SOC, and static time will affect the test results of EIS, so it is necessary to test under the same conditions to eliminate the interference of external factors. In addition, the frequency scanning range and scanning points need to be set reasonably to take into account both test accuracy and efficiency. The frequency scanning range usually covers the mHz to kHz level, and the scanning points are generally set to 50~100 to obtain sufficient frequency resolution.

[0040] The EIS data obtained in step S1220 lays the foundation for subsequent feature extraction and health status assessment. EIS data not only reflects the complex electrochemical process inside the battery cell, but also reveals the evolution of the battery cell aging state. By tracking EIS data at different life cycle stages, dynamic monitoring of the health status of the battery cell can be achieved, and safety hazards of the battery cell can be discovered and warned in a timely manner. Based on the analysis of EIS data, we can deeply understand the aging mechanism of the battery cell, optimize the battery management strategy, and extend the service life of the battery cell.

[0041] The EIS feature matrix is ​​a compact and efficient data representation that can condense the rich information contained in the EIS data into several key parameters, which is convenient for subsequent feature fusion and health status assessment. In step S1231, the electrochemical impedance spectroscopy data is fitted with an equivalent circuit to obtain the equivalent circuit model parameters. The equivalent circuit model is a commonly used EIS data analysis method. By constructing a circuit model corresponding to the battery characteristics, a quantitative interpretation of the EIS data is achieved. Commonly used equivalent circuit models include Randles circuit, Thevenin circuit, etc. Among them, the Randles circuit is composed of components such as ohmic internal resistance, double layer capacitance, charge transfer resistance and Warburg impedance, which can well describe the electrochemical properties of the electrode / electrolyte interface. The Thevenin circuit is composed of components such as ohmic internal resistance, polarization resistance and polarization capacitance, which can characterize the macroscopic characteristics of the battery. By selecting a suitable equivalent circuit model and fitting the EIS data using algorithms such as nonlinear least squares method, the key parameters in the equivalent circuit can be obtained to quantitatively describe the internal characteristics of the battery.

[0042] In step S1232, key impedance parameters are extracted from the equivalent circuit model parameters. Different impedance parameters reflect the characteristics of different electrochemical processes inside the battery cell. For example, the charge transfer resistance reflects the resistance to charge transfer at the electrode / electrolyte interface, which is closely related to the electrode reaction kinetics; the double layer capacitance reflects the polarization characteristics of the electrode / electrolyte interface, which is related to the electrode specific surface area and roughness; the Warburg impedance reflects the diffusion characteristics of ions in the electrolyte, which is related to the electrolyte concentration and diffusion coefficient. By extracting these key impedance parameters, the electrochemical characteristics inside the battery cell can be quantitatively evaluated, revealing the degradation mechanism of the battery cell health state.

[0043] In step S1233, an EIS feature matrix is ​​constructed from key impedance parameters. The EIS feature matrix is ​​a multidimensional data representation, in which each row corresponds to an EIS test sample and each column corresponds to a key impedance parameter. By arranging the impedance parameters of different samples in the same matrix, data comparison and analysis can be conveniently performed. In order to eliminate the influence of dimensional differences, the EIS feature matrix also needs to be standardized to unify the value range of each parameter to the interval [0,1] or [-1,1]. Standardization can improve the comparability between different parameters and facilitate subsequent feature selection and fusion.

[0044] Through step S1230, the present invention realizes the efficient conversion from EIS data to EIS characteristic matrix. The EIS characteristic matrix not only simplifies the representation of EIS data, but also condenses the key information in the EIS data. By extracting key parameters such as charge transfer resistance, double layer capacitance, Warburg impedance, etc., the EIS characteristic matrix quantitatively depicts the complex electrochemical properties inside the battery cell, laying the foundation for subsequent multi-parameter fusion and health status assessment. Based on the analysis of the EIS characteristic matrix, the aging behaviors such as battery cell capacity attenuation, power drop, impedance increase, etc. can be quantitatively evaluated, revealing the laws of battery cell health evolution, and providing a reliable basis for fault diagnosis and life prediction.

[0045] Step S1300, integrating the nonlinear frequency response feature matrix and the EIS feature matrix to construct a refined multi-dimensional electrochemical feature space; based on the refined multi-dimensional electrochemical feature space, constructing and training a battery cell health assessment model, and extracting battery cell health indicators; establishing a battery cell thermal runaway risk level, and constructing a mapping relationship between the battery cell health indicators and the battery cell thermal runaway risk level.

[0046] Further, step S1300 includes: Step S1310, fusing the nonlinear frequency response characteristic matrix and the EIS characteristic matrix to construct a refined multi-dimensional electrochemical characteristic space; Furthermore, if Figure 5 As shown, step S1310 includes: Step S1311, performing data standardization processing on the nonlinear frequency response characteristic matrix and the EIS characteristic matrix; Step S1312, splicing the standardized nonlinear frequency response characteristic matrix and the EIS characteristic matrix by column to form a multidimensional electrochemical characteristic matrix; each row of the multidimensional electrochemical characteristic matrix corresponds to a battery cell sample, and each column corresponds to an electrochemical characteristic; Step S1313, performing feature selection on the multi-dimensional electrochemical feature matrix; Step S1314, performing feature extraction on the multidimensional electrochemical feature matrix after feature selection to obtain a refined multidimensional electrochemical feature space, wherein the refined multidimensional electrochemical feature space contains n1 electrochemical features.

[0047] Specifically, the refined multi-dimensional electrochemical feature space is a highly condensed data representation that integrates multiple electrochemical characteristics of the battery cell and comprehensively describes the health status of the battery cell. By integrating nonlinear frequency response characteristics and EIS characteristics, the refined multi-dimensional electrochemical feature space fully explores the inherent laws contained in thermal-electric coupling data and impedance spectrum data, laying a solid data foundation for subsequent health assessment.

[0048] Data standardization is a commonly used data preprocessing method. It can unify feature data of different dimensions to the same scale and eliminate the impact of dimension differences on subsequent analysis. Common data standardization methods include maximum-minimum standardization, zero-mean standardization, and decimal scaling standardization. Among them, maximum-minimum standardization linearly scales feature data to the interval [0,1], zero-mean standardization scales feature data to the interval with a mean of 0 and a standard deviation of 1, and decimal scaling standardization scales feature data to the interval [-1,1]. Through data standardization, the comparability between different electrochemical features can be improved, which is convenient for subsequent feature fusion and analysis. The multidimensional electrochemical feature matrix is ​​a high-dimensional data representation that organically combines different types of electrochemical features and comprehensively describes the health status of the battery cell. By splicing by columns, the multidimensional electrochemical feature matrix integrates nonlinear frequency response features and EIS features into the same matrix, with each row corresponding to a battery cell sample and each column corresponding to an electrochemical feature. This compact data organization form not only facilitates subsequent feature selection and extraction, but also provides convenience for building a health assessment model.

[0049] Feature selection is a commonly used dimensionality reduction method. It selects the most representative and discriminative feature subsets by evaluating the relevance, redundancy and predictive ability of features. Common feature selection methods include filtering, wrapping, embedding, etc. Among them, the filtering method independently evaluates the importance of each feature based on the statistical indicators of the feature (such as variance, correlation coefficient, etc.), the wrapping method combines feature selection with the learning algorithm, and obtains the optimal feature subset through iterative search, and the embedding method embeds feature selection into the learning algorithm, and realizes feature selection and model training simultaneously by optimizing the objective function. Through feature selection, redundant and irrelevant features can be removed, the feature dimension can be reduced, and the efficiency and accuracy of subsequent health assessment can be improved. Feature extraction maps the original high-dimensional feature space to a low-dimensional subspace through a certain transformation, and realizes data compression and noise reduction while retaining the intrinsic structure of the data. Common feature extraction methods include principal component analysis (PCA), linear discriminant analysis (LDA), independent component analysis (ICA), etc. Among them, PCA converts the original features into linearly independent principal components through orthogonal transformation and sorts them according to the size of variance. The first few principal components usually carry most of the information; LDA maximizes the divergence between classes and minimizes the divergence within classes to find the optimal projection direction to achieve the maximum separation of data of different categories; ICA maximizes the statistical independence between components to find independent representations of data. Through feature extraction, the data dimension can be greatly reduced while retaining the essential structure of the data, resulting in a refined multidimensional electrochemical feature space. This feature space usually contains n1 electrochemical features, which retain the information of the original data to the greatest extent and provide high-quality input for subsequent health assessment.

[0050] Through step S1310, the present invention constructs a refined multi-dimensional electrochemical feature space, and realizes a comprehensive characterization of the health status of the battery cell. The refined multi-dimensional electrochemical feature space integrates nonlinear frequency response characteristics and EIS characteristics, fully explores the inherent laws contained in the thermal-electric coupling data and impedance spectrum data, and provides rich feature information for health assessment. Through data standardization processing, the present invention eliminates the dimensional differences between different electrochemical characteristics and improves the comparability of characteristics; through column splicing, the present invention integrates different types of electrochemical characteristics into the same mathematical framework and constructs a multi-dimensional electrochemical feature matrix; through feature selection and extraction, the present invention removes redundant and irrelevant features, while retaining the intrinsic structure of the data, greatly reducing the data dimension, and obtaining a refined multi-dimensional electrochemical feature space. Compared with the original thermal-electric coupling data and impedance spectrum data, the refined multi-dimensional electrochemical feature space has higher information density and stronger characterization capabilities, laying a solid data foundation for subsequent health assessment.

[0051] For example, suppose that the nonlinear frequency response characteristic matrix and EIS characteristic matrix of 100 battery cell samples are obtained, and each matrix contains 50 features. In sub-step S1311, the two matrices are respectively subjected to maximum-minimum value normalization processing, and all feature data are scaled to the interval [0,1]. In sub-step S1312, the two standardized matrices are spliced ​​by column to form a 100×100 multidimensional electrochemical feature matrix. In sub-step S1313, the multidimensional electrochemical feature matrix is ​​selected by filtering method, and the 30 most representative electrochemical features are selected according to the variance and correlation coefficient of the features. In sub-step S1314, PCA is used to extract features from the matrix after feature selection, and the 30-dimensional feature space is compressed into a 10-dimensional refined multidimensional electrochemical feature space. By analyzing the refined multidimensional electrochemical feature space, it is found that battery cell samples in different health states show obvious clustering characteristics in this space, indicating that this feature space has a strong characterization ability for the health state of the battery cell. Based on this discovery, an efficient and accurate health assessment model was constructed to achieve quantitative assessment and prediction of the health status of battery cells.

[0052] In short, the refined multidimensional electrochemical feature space constructed by step S1310 and its sub-steps has opened up a new way for the assessment of the health status of battery cells. By integrating nonlinear frequency response characteristics and EIS characteristics, and standardizing, selecting and extracting characteristic data, the refined multidimensional electrochemical feature space retains the inherent laws of battery cell health degradation to the greatest extent, and provides high-quality input for the construction of a health assessment model. The introduction of this feature space not only expands the analysis dimension of battery cell characterization data, but also provides new ideas for multivariate data fusion. In future research, we can further explore the integration of refined multidimensional electrochemical feature space with other physical and chemical characteristics to build a more comprehensive and accurate battery cell health assessment system.

[0053] Step S1320, constructing and training a cell health assessment model based on the refined multi-dimensional electrochemical feature space.

[0054] Furthermore, if Figure 6 As shown, step S1320 includes: Step S1321, constructing a training sample set using the electrochemical features in the refined multi-dimensional electrochemical feature space as sample features and the manually labeled cell health indicators as sample labels; Step S1322, using a weighted linear regression model to construct an initial cell health assessment model; Step S1323, using the training sample set to train the initial cell health assessment model to obtain the weight coefficient and bias term b of each electrochemical feature; Step S1324, based on the weight coefficient of each electrochemical feature and the bias term b, a final cell health assessment model is obtained.

[0055] Specifically, the battery cell health assessment model is a mathematical model for quantitatively assessing the health status of the battery cell. It achieves quantitative prediction of the health status of the battery cell by establishing a mapping relationship between the refined multi-dimensional electrochemical characteristics and the battery cell health indicators. The construction and training of the battery cell health assessment model is a data-driven process. It needs to be based on a large amount of experimental data and prior knowledge, and use intelligent algorithms such as machine learning to mine the inherent laws of battery cell health degradation from the data. The training sample set is the basis of model training. It consists of a series of input-output data pairs, reflecting the correspondence between electrochemical characteristics and health indicators. Among them, the sample features refer to the electrochemical features in the refined multi-dimensional electrochemical feature space, including the fused nonlinear frequency response features and EIS features, which comprehensively characterize the health status of the battery cell; the sample labels refer to the manually labeled battery cell health indicators, such as capacity retention rate, internal resistance increase rate, etc., which quantitatively reflect the health level of the battery cell. By collecting a certain number of battery cell samples and performing standardized charge and discharge tests and electrochemical characterization on each sample, a comprehensive and reliable training sample set can be obtained. The quality of the sample set directly determines the training effect and generalization performance of the model. Therefore, it is necessary to adopt a reasonable experimental design and data collection plan to cover as many battery cell samples as possible under different working conditions, with different materials, and from different production batches, so as to improve the representativeness and diversity of the sample set.

[0056] Weighted linear regression establishes a linear relationship between independent variables and dependent variables by minimizing the sum of squared errors. In the present invention, the weighted linear regression model takes the refined multidimensional electrochemical characteristics as independent variables and the battery cell health index as the dependent variable, and constructs an initial health assessment model through linear combination. This model has the advantages of simple form and efficient calculation, and can quickly give a quantitative estimate of the health status of the battery cell. It should be noted that due to the different contributions of different electrochemical characteristics to the health index, it is necessary to introduce weight coefficients to adjust the importance of each feature. The size of the weight coefficient reflects the relative importance of the corresponding feature, which can be optimized by empirical setting or data-driven method. In addition, it is necessary to set a bias term to correct the overall deviation of the model. By reasonably setting the weight coefficient and bias term, a preliminary battery cell health assessment model can be obtained, which provides a basis for subsequent model optimization.

[0057] Model training is a process of parameter optimization. By minimizing the error between the model prediction value and the true value, the model parameters are continuously adjusted so that the model can fit the training data as accurately as possible. In the present invention, the training sample set is used to train and optimize the initial battery cell health assessment model. Specifically, the training sample set is input into the model, the predicted value of the model is calculated, and compared with the sample label to obtain the prediction error. Then, an optimization algorithm such as gradient descent is used to adjust the model parameters based on the prediction error to minimize the error. Through multiple iterative optimizations, a set of optimal weight coefficients and bias terms can be obtained, so that the model can fit the training data to the greatest extent. These weight coefficients quantify the degree of influence of different electrochemical characteristics on the health index. The larger the value, the more important the corresponding feature; the bias term characterizes the overall deviation of the model and is used to correct the systematic error of the model. In the process of model training, it is necessary to use methods such as cross-validation to reasonably divide the training set and the validation set to avoid the occurrence of overfitting problems. Through sufficient training and verification, a battery cell health assessment model with excellent performance can be obtained, which lays the foundation for subsequent health prediction and thermal runaway warning.

[0058] Through the model training of step S1323, a set of optimal weight coefficients and bias terms are obtained, which characterize the quantitative relationship between the refined multidimensional electrochemical characteristics and the battery cell health index. Substituting these parameters into the initial weighted linear regression model, the final battery cell health assessment model is obtained. The model takes the refined multidimensional electrochemical characteristics as input and calculates the quantitative prediction value of the battery cell health index by weighted linear combination. The higher the prediction value, the better the health status of the battery cell; the lower the prediction value, the worse the health status of the battery cell, and there is a greater risk of thermal runaway. The performance of the health assessment model directly determines the accuracy and reliability of the prediction results. In order to comprehensively evaluate the performance of the model, it is necessary to verify the model on an independent test sample set. By comparing the error between the model prediction value and the true value, calculating the mean square error, absolute error and other evaluation indicators, the prediction accuracy and robustness of the model can be quantitatively evaluated. A good health assessment model should be able to give accurate and consistent health prediction results for battery cell samples under different operating conditions, with different materials and from different production batches, thus providing a reliable decision-making basis for the battery management system.

[0059] Through step S1320 and its sub-steps, the present invention constructs and trains an efficient and accurate battery cell health assessment model. The model takes the refined multi-dimensional electrochemical features as input, fully mines the health degradation information contained in the electrochemical characterization data, overcomes the limitations of a single feature, and achieves a comprehensive characterization of the health status of the battery cell. Through machine learning algorithms such as weighted linear regression, the model establishes a quantitative relationship between electrochemical characteristics and health indicators, and can quickly and accurately predict the health status of the battery cell based on its real-time electrochemical performance, providing reliable data support for thermal runaway warning. Compared with the traditional single indicator evaluation method, the health assessment model of the present invention has stronger robustness and adaptability, can be applied to batteries of different types and under different usage conditions, and greatly improves the efficiency and accuracy of health management.

[0060] For example, assume that the performance data of 100 battery cell samples are collected, and each sample is subjected to standardized charge and discharge tests and EIS characterization. Through step S1310, the nonlinear frequency response characteristics and EIS characteristics are integrated to obtain a 10-dimensional refined multidimensional electrochemical feature space. At the same time, the health index of each sample is obtained through tests such as capacity and internal resistance, and is manually labeled. In step S1321, the refined multidimensional electrochemical characteristics are used as sample features, and the manually labeled health indicators are used as sample labels to construct a training set containing 100 samples. In step S1322, a weighted linear regression model is used to construct an initial health assessment model, in which the initial values ​​of the weight coefficients and bias terms are randomly set. In step S1323, the initial model is trained and optimized using the training set, and a set of optimal weight coefficients and bias terms are obtained through algorithms such as the least squares method. For example, it is found that the weight coefficients of charge transfer resistance and double-layer capacitance are the largest, indicating that they have the most significant impact on the health status of the battery cell. In step S1324, the optimal parameters are substituted into the initial model to obtain the final battery cell health assessment model. In order to evaluate the performance of the model, 20 independent battery cell samples were collected, subjected to the same test and characterization, and predicted using the trained model. The results show that the model can predict the health indicators of the battery cell with an accuracy of 95%, and the mean error is less than 5%. This proves that the model has good predictive performance and practical value, and can be applied to actual battery health management systems.

[0061] In summary, step S1320 and its sub-steps construct and train an efficient and accurate battery cell health assessment model, providing key technical support for battery health management and thermal runaway warning. The model makes full use of multivariate electrochemical characterization data, and through intelligent algorithms such as machine learning, it explores the inherent laws of battery cell health degradation and realizes quantitative assessment of health status. The model construction ideas and training methods can not only be extended to other types of battery systems, but also provide new solutions to health assessment problems in other engineering fields.

[0062] Step S1330, establishing a thermal runaway risk level for a battery cell, and constructing a mapping relationship between a battery cell health index and a thermal runaway risk level for a battery cell.

[0063] Further, step S1330 includes: Step S1331, obtaining the material characteristics and design parameters of the battery cell, analyzing the thermal runaway mechanism of the battery cell, and determining the key factors affecting the thermal runaway; Step S1332, classifying the thermal runaway risk level; Step S1333: Based on the key factors affecting thermal runaway and the risk level classification, a mapping relationship between the battery cell health index and the thermal runaway risk level is established.

[0064] Specifically, the thermal runaway risk level is a qualitative risk assessment indicator, which divides the thermal runaway risk of the battery cell into several levels, such as low risk, medium risk, high risk, etc., reflecting the possibility and severity of thermal runaway of the battery cell. By establishing a mapping relationship between the health index and the thermal runaway risk level, the quantitative health assessment results can be converted into qualitative risk warning information, providing an intuitive and operational decision-making basis for the battery management system. The thermal runaway mechanism refers to a series of physical and chemical processes of heat accumulation and temperature runaway that occur in the battery cell under specific conditions, such as decomposition of positive electrode materials, flammability of electrolytes, and melting of diaphragms. By analyzing the material properties of the battery cell (such as the thermal stability of positive and negative electrode materials, the flammability of electrolytes, etc.) and design parameters (such as the energy density of the battery cell, the heat dissipation structure, etc.), the root cause of thermal runaway can be revealed from a mechanism perspective, and the key factors affecting thermal runaway can be determined. These key factors may include operating parameters such as battery cell temperature, SOC, and charge and discharge rate, and may also include health indicators such as battery cell capacity and internal resistance. The analysis of thermal runaway mechanism is the theoretical basis for risk level classification, and it provides a scientific basis for subsequent risk judgment and early warning.

[0065] Based on the results of the thermal runaway mechanism analysis, combined with the safety margin of the battery cell and the actual operating conditions, the thermal runaway risk is divided into several levels. Common risk level classification methods include the three-level method (low risk, medium risk, high risk), the five-level method (very low risk, low risk, medium risk, high risk, very high risk), etc. The classification of risk levels needs to take into account both safety and practicality. It must be detailed enough to reflect the difference in risks, and simple enough to facilitate operation. When dividing the risk levels, it is necessary to formulate scientific and reasonable classification thresholds to clarify the risk range and judgment criteria of each level. The formulation of these thresholds needs to comprehensively consider factors such as the material properties, design parameters, operating conditions of the battery cell, as well as external factors such as industry standards, regulations and policies. By reasonably dividing the risk levels, a basis can be provided for subsequent risk warnings and decision-making.

[0066] The mapping relationship is a quantitative to qualitative transformation method, which maps continuous health indicators to discrete risk levels, realizing the qualitative description of risks. The key to establishing a mapping relationship is to find the corresponding law between health indicators and risk levels, that is, at what health threshold, the thermal runaway risk of the battery cell reaches what level. This corresponding law can be obtained through a large amount of experimental data and statistical analysis, or it can be set according to the thermal runaway mechanism and expert experience. For example, the health indicator can be divided into several intervals according to the safety thresholds of parameters such as temperature and SOC, and each interval corresponds to a risk level; fuzzy logic, decision tree and other methods can also be used to establish a nonlinear mapping relationship between health indicators and risk levels. By reasonably constructing a mapping relationship, the health degradation process of the battery cell can be associated with the evolution process of the thermal runaway risk, realizing dynamic risk assessment and early warning.

[0067] Through step S1330, the present invention establishes the thermal runaway risk level of the battery cell and constructs a mapping relationship between the battery cell health index and the battery cell thermal runaway risk level. The division of thermal runaway risk levels is based on the analysis of the thermal runaway mechanism, fully considering the influence of factors such as battery cell materials, design, and operating conditions, and reasonably setting the risk classification threshold, which can accurately reflect the thermal runaway risk level of the battery cell. The mapping relationship between the health index and the risk level quantifies the intrinsic connection between the two and realizes the quantitative to qualitative transformation of risk. By timely evaluating the health index of the battery cell and judging its risk level based on the mapping relationship, real-time monitoring and early warning of thermal runaway risks can be achieved, providing reliable decision-making information for the battery management system.

[0068] Thermal runaway risk level and health-risk mapping is an innovative risk management method that makes up for the shortcomings of traditional methods and realizes the quantification, dynamicization and visualization of risk assessment. Traditional risk management methods are often based on a single indicator (such as temperature, voltage, etc.), lack of globality and systematicity, and cannot accurately grasp the entire process of battery cell health degradation. The method of the present invention is based on the health index, comprehensively considers the influence of multi-source heterogeneous data, and comprehensively describes the evolution law of the battery cell state, and its risk judgment is more scientific and accurate. In addition, the mapping of health indicators and risk levels also makes the risk assessment results more intuitive and operational, which is conducive to the rapid decision-making and feedback control of the battery management system.

[0069] For example, if the analysis of the thermal runaway mechanism of a certain battery cell shows that when the temperature exceeds 150°C and the SOC is higher than 90%, the positive electrode material is very likely to decompose and the electrolyte burns, and the risk of thermal runaway is extremely high. Based on this, the thermal runaway risk level can be reasonably divided, such as: low risk (temperature below 100°C, SOC below 70%), medium risk (temperature within 100-150°C, SOC within 70-90%), and high risk (temperature above 150°C, SOC above 90%). On this basis, a mapping relationship between health indicators and risk levels is established, such as: health indicators above 0.8 are low risks; between 0.6-0.8 are medium risks; below 0.6 are high risks. In this way, when the health index of a battery cell is evaluated as 0.75, it can be judged that it is at a medium-risk level and certain preventive measures need to be taken (such as reducing the charging rate, enhancing heat dissipation, etc.); when the health index drops to 0.55, it indicates that it is at a high-risk level and needs to be stopped immediately to avoid thermal runaway accidents.

[0070] In short, the thermal runaway risk level and health-risk mapping established in step S1330 is an advanced battery safety management technology. It is based on the analysis of thermal runaway mechanism and uses health index as a bridge to closely link the health degradation process of the battery cell with the evolution process of thermal runaway risk, thus realizing real-time monitoring, quantitative early warning and dynamic control of risks.

[0071] Step S2000, obtaining the real-time operating condition parameters of the battery cell, and adaptively adjusting the weight coefficients of each electrochemical characteristic in the battery cell health assessment model according to the real-time operating condition parameters of the battery cell to obtain an operating condition adaptive battery cell health index; inputting the operating condition adaptive battery cell health index into a pre-established mapping relationship between the battery cell health index and the thermal runaway risk level to obtain a predicted thermal runaway risk level; determining whether the predicted thermal runaway risk level exceeds a preset threshold, and if so, triggering a warning mechanism.

[0072] Furthermore, step S2000 includes: Step S2100, obtaining the real-time operating condition parameters of the battery cell, and adaptively adjusting the weight coefficients of the electrochemical characteristics in the battery cell health assessment model according to the real-time operating condition parameters of the battery cell to obtain the operating condition adaptive battery cell health index; Furthermore, step S2100 includes: Step S2110, collecting real-time operating parameters of the battery cell, wherein the real-time operating parameters of the battery cell include ambient temperature, charge and discharge rate, and battery cell SOC; Step S2120, dynamically adjust the weight coefficient of each electrochemical characteristic in the battery cell health assessment model in combination with the real-time operating condition parameters of the battery cell to obtain an operating condition adaptive battery cell health index.

[0073] Furthermore, if Figure 7 As shown, step S2120 includes: Step S2121, obtaining the operating parameters of the battery cell, and constructing a mapping relationship between the operating parameters and the weight adjustment coefficient; Step S2122, calculating the real-time weight adjustment coefficient according to the real-time operating parameters of the battery cell and the mapping relationship between the operating parameters and the weight adjustment coefficient; Step S2123, adjusting the weight coefficient of each electrochemical feature in the battery cell health assessment model according to the real-time weight adjustment coefficient to obtain an adjusted weight coefficient; Step S2124, using the adjusted weight coefficient, calculate the operating condition adaptive battery cell health index under the current operating condition.

[0074] Specifically, real-time operating parameters refer to the dynamic parameters of the battery cell during actual use. They reflect the working environment and state of the battery cell, and have an important impact on the health degradation and thermal runaway risk of the battery cell. By collecting these parameters in real time, the real-time state of the battery cell can be mastered, providing important input information for subsequent health assessment and risk warning. Among them, the ambient temperature refers to the temperature of the environment in which the battery cell is located, which directly affects the electrochemical reaction rate and thermal equilibrium conditions inside the battery cell. Too high ambient temperature will accelerate the capacity decay and impedance increase of the battery cell, increasing the risk of thermal runaway; too low ambient temperature will reduce the charging and discharging efficiency and energy utilization of the battery cell, affecting the working performance of the battery cell. Therefore, real-time monitoring of ambient temperature is crucial for evaluating the health status of the battery cell and preventing thermal runaway. Commonly used ambient temperature sensors include thermocouples, thermal resistors, thermistors, etc. By reasonably arranging the sensor positions, the temperature distribution information of the environment in which the battery cell is located can be obtained. The charge and discharge rate refers to the ratio of the charge and discharge current of the battery cell to the rated capacity, which reflects the charge and discharge intensity and rate of the battery cell. If the charge and discharge rate is too high, the polarization effect and thermal effect inside the battery cell will be aggravated, resulting in a rapid rise in the battery cell temperature and an increased risk of thermal runaway; if the charge and discharge rate is too low, the charge and discharge time will be prolonged and the efficiency of the battery cell will be reduced. Therefore, real-time monitoring of the charge and discharge rate is very necessary for adjusting the charge and discharge strategy of the battery cell and preventing thermal runaway. The charge and discharge rate can be calculated by measuring the charge and discharge current and rated capacity of the battery cell, or it can be directly obtained through the battery management system. If the SOC is too high, the redox reaction activity of the battery cell will be increased, the decomposition of the electrolyte and the release of gas will be aggravated, and the risk of thermal runaway will be increased; if the SOC is too low, the over-discharge of the battery cell will be deepened, resulting in structural damage and capacity attenuation of the electrode material. Therefore, real-time monitoring of the SOC is crucial for optimizing the charge and discharge range of the battery cell and preventing thermal runaway. The SOC can be estimated by methods such as the Coulomb counting method, the open circuit voltage method, and the Kalman filter method, or it can be directly obtained through the battery management system. The real-time operating parameters of the battery cell collected in step S2110 provide important decision-making information for subsequent health assessment and risk warning. On the one hand, these parameters reflect the actual working state of the battery cell and affect the rate and mode of health degradation of the battery cell; on the other hand, they are closely related to the triggering conditions of thermal runaway of the battery cell and affect the level of thermal runaway risk. By collecting and analyzing these parameters in real time, the battery cell health assessment model can be dynamically adjusted to improve the adaptability and accuracy of health assessment; at the same time, abnormal changes in thermal runaway risks can be discovered in a timely manner, triggering corresponding early warning mechanisms and safety protection measures.

[0075] The condition-adaptive cell health index is a dynamic evaluation index that takes into account the impact of the actual working conditions of the cell on the health status, and can more accurately and timely reflect the health degradation of the cell. By obtaining the working condition parameters in real time and dynamically adjusting the health assessment model accordingly, the adaptability and robustness of the health assessment can be significantly improved, providing reliable decision support for subsequent risk warnings and safety protection.

[0076] The weight adjustment coefficient is an intermediate variable that quantifies the impact of the working condition. It reflects the contribution of electrochemical characteristics to the health status of the battery cell under different working conditions. By establishing a mapping relationship between working condition parameters and weight adjustment coefficients, the adjustment law of the weight of electrochemical characteristics due to working condition changes can be quantitatively described. This mapping relationship can be constructed based on mechanism analysis, expert experience or data-driven methods. For example, through theoretical analysis, the quantitative relationship between parameters such as ambient temperature and charge and discharge rate and processes such as electrochemical reaction rate and mass transfer and heat transfer can be derived to determine their influence on the weight of electrochemical characteristics; it is also possible to fit the nonlinear mapping relationship between working condition parameters and weight adjustment coefficients through a large amount of experimental data and machine learning methods such as neural networks and support vector machines. The real-time weight adjustment coefficient is a set of dynamically changing quantitative indicators that reflect the relative importance of the weights of various electrochemical characteristics under the current working conditions in real time. By substituting the real-time collected working condition parameters into the pre-constructed mapping relationship, the weight adjustment coefficients corresponding to each working condition parameter can be quickly calculated. It should be noted that since there may be certain measurement errors and random noise in the operating parameters, in order to improve the robustness of the calculation, certain data smoothing and filtering methods, such as moving average method, Kalman filtering method, etc., can be used to pre-process the original operating data to reduce the impact of errors and noise.

[0077] The adjusted weight coefficients are a set of dynamically updated model parameters that comprehensively consider the dual effects of operating conditions and electrochemical characteristics, and can more accurately reflect the changing laws of the health status of the battery cell. The specific method of adjustment can be to use mathematical operations such as weighted average and product scaling to combine the original weight coefficients with the corresponding adjustment coefficients to obtain new weight coefficients. It should be noted that the adjusted weight coefficients must be normalized to ensure that the value range and proportional relationship of each weight coefficient meet the constraints of the model. The operating condition adaptive battery cell health index is a comprehensive evaluation index that quantifies the health status of the battery cell. It fully considers the influence of operating conditions and electrochemical characteristics, and can dynamically reflect the instantaneous state of health degradation of the battery cell. The specific method of calculation is to substitute the electrochemical characteristic data collected in real time into the battery cell health assessment model, and use the adjusted weight coefficients for weighted combination to obtain a scalar value as the health indicator at the current moment. Compared with the traditional static health indicator, the operating condition adaptive battery cell health indicator has higher timeliness and pertinence, and can provide a more dynamic and accurate health assessment result for the battery management system.

[0078] Through step S2120, the present invention realizes the working condition adaptive adjustment of the battery cell health assessment model, and obtains a new working condition adaptive battery cell health index. This indicator makes full use of the real-time working condition information of the battery cell, dynamically adjusts the health assessment model by quantifying the relationship between the working condition parameters and the electrochemical characteristic weights, and realizes the real-time update and adaptive optimization of the health status assessment. Compared with the traditional static health assessment method, the working condition adaptive battery cell health index can capture the dynamic changes of the health status of the battery cell more accurately and timely, and provide a more reliable and effective basis for the regulation and optimization of the battery management system.

[0079] The introduction of the condition-adaptive cell health index also provides new research ideas and directions for the field of battery health management. Traditional battery health assessment methods are often based on offline, static models and data, which are difficult to adapt to complex and changeable working conditions, and the assessment results are difficult to reflect the real-time dynamics of the health status of the battery cells. The condition-adaptive health assessment method proposed in the present invention realizes real-time synchronization and adaptive optimization of the health assessment model and the cell working conditions by real-time perception of the working condition parameters and dynamic adjustment of the model weights, providing a new technical means for realizing health monitoring and management of the entire life cycle of the battery.

[0080] Step S2200, inputting the working condition adaptive battery cell health index into the mapping relationship between the battery cell health index and the thermal runaway risk level to obtain a predicted thermal runaway risk level; Specifically, the condition-adaptive battery cell health index is obtained in step S2100. It comprehensively considers the influence of the real-time operating parameters of the battery cell, and adaptively adjusts the battery cell health assessment model, which can more accurately reflect the actual health status of the battery cell under the current operating conditions. The mapping relationship between the battery cell health index and the thermal runaway risk level is established in step S1330. It quantitatively describes the influence of health deterioration on thermal runaway risk and provides a quantitative basis for risk warning. By inputting the condition-adaptive battery cell health index into the health-risk mapping relationship, a predicted thermal runaway risk level can be obtained. This predicted risk level reflects the possibility and severity of thermal runaway of the battery cell under the current health state and operating conditions, and is a qualitative risk description indicator. According to the predicted risk level, the safety status of the battery cell can be preliminarily judged, providing a basis for subsequent risk warning and safety protection.

[0081] Step S2200 organically combines the condition-adaptive health assessment with the qualitative risk level warning, realizing real-time monitoring of the safety status of the battery cell and risk warning. Compared with the traditional threshold warning method based on a single indicator, this method can comprehensively consider the influence of the health status of the battery cell and the operating conditions, and give a more comprehensive and accurate risk prediction result. This prediction result not only reflects the degree of health degradation of the battery cell, but also takes into account the influence of the operating conditions, and can be closer to the actual use of the battery cell. Safety protection measures based on this prediction result will also be more targeted and effective, which can minimize the risk of thermal runaway and ensure the safe operation of the battery.

[0082] For example, suppose that in step S2100, the condition-adaptive cell health index of a cell at a certain moment is 0.75. The index is input into the pre-established health-risk mapping relationship, and the predicted thermal runaway risk level of the cell in the current state is obtained as "medium risk". This prediction result shows that although the overall health status of the cell is acceptable, under the current operating conditions (such as high ambient temperature, large charge and discharge rate, etc.), its thermal runaway risk has reached a high level, and certain preventive measures need to be taken, such as reducing the charge and discharge rate, enhancing heat dissipation, etc. Without these preventive measures, the cell continues to be used in a high-risk state, and its thermal runaway risk will continue to rise, which may cause safety accidents such as fire and explosion in severe cases. Therefore, timely warning and intervention are crucial to ensure battery safety.

[0083] In summary, step S2200 realizes the real-time prediction and early warning of the thermal runaway risk of the battery cell by combining the working condition adaptive battery cell health index with the thermal runaway risk level mapping. This method makes full use of the health assessment results and risk level classification results obtained in the previous steps, organically unifies the quantitative health assessment with the qualitative risk level warning, and forms a complete battery cell safety status monitoring and risk prevention and control system. The system can dynamically reflect the changes in the health status and safety risks of the battery cell, provide a reliable decision-making basis for the battery management system, and maximize the safe operation of the battery. At the same time, the health assessment and risk warning models based on this method are all built based on data-driven machine learning methods, and have strong adaptability and extensibility. By continuously updating and optimizing the training data, the model can adapt to different types of battery cells and working conditions, and its prediction performance will continue to improve.

[0084] Step S2300, determine whether the predicted thermal runaway risk level exceeds a preset threshold; if it exceeds the preset threshold, trigger the early warning mechanism in time and take corresponding safety protection measures.

[0085] Specifically, the preset threshold is reasonably set based on factors such as the cell material characteristics, design parameters, and safety margins, and it defines the safety boundary of the cell thermal runaway risk. When the predicted risk level exceeds this threshold, it means that the cell has entered a high-risk state. If timely measures are not taken, the occurrence of a thermal runaway accident will be a high-probability event, and the consequences will be catastrophic. Therefore, timely triggering of the early warning mechanism and taking effective safety protection measures are crucial to avoiding accidents and ensuring battery safety.

[0086] The setting of preset thresholds requires comprehensive consideration of many factors. First, the thermal stability and safety performance of the battery material need to be considered. The thermal runaway trigger temperature, heat release rate and other characteristics of batteries with different material systems vary greatly, which directly affects the level of thermal runaway risk. Therefore, the setting of preset thresholds needs to give specific recommended values ​​for different battery materials. Secondly, the design parameters of the battery, such as energy density and heat dissipation structure, need to be considered. The design parameters affect the temperature distribution and heat accumulation inside the battery, which in turn affects the risk of thermal runaway. For different design parameters, the preset threshold also needs to be adjusted accordingly. Thirdly, the safety margin needs to be considered. The size of the safety margin determines the sensitivity and reliability of the early warning mechanism. If the margin is set too large, it may lead to too many false alarms; if the margin is set too small, risks may not be discovered in time. Therefore, it is necessary to give a balanced threshold recommendation based on a comprehensive assessment of the safety performance of the battery.

[0087] When the predicted thermal runaway risk level exceeds the preset threshold, the early warning mechanism will be automatically triggered. The early warning information will be promptly transmitted to the battery management system through various channels (such as display screens, sound and light alarms, communication networks, etc.), and then take corresponding safety protection measures according to the level of the early warning level and its own functional positioning. These measures may include: reducing the charging and discharging current, increasing the power of the cooling fan, preparing fire extinguishing equipment, arranging personnel evacuation, etc. Through these measures, the risk of thermal runaway of the battery cell can be effectively reduced, the further deterioration of the accident can be avoided, and the safety of personnel and equipment can be protected.

[0088] In summary, step S2300 sets a reasonable warning threshold and takes safety protection measures in time when the predicted risk exceeds the standard, thus building a complete set of thermal runaway risk warning and prevention and control mechanisms for battery cells. This mechanism is based on the risk level prediction results of step S2200, and fully considers the influence of factors such as battery cell materials, design parameters, and engineering applications. It can effectively identify and control the thermal runaway risks of battery cells and improve battery safety and reliability.

[0089] Example 2 This embodiment provides a multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells on the basis of Embodiment 1, including: Step S1313, performing feature selection on the multi-dimensional electrochemical feature matrix; Furthermore, if Figure 8 As shown, step S1313 includes: Step S13131, calculating the importance score of each electrochemical feature in the multidimensional electrochemical feature matrix; Step S13132, constructing an electrochemical feature importance matrix based on the importance score of each electrochemical feature; Step S13133, based on the electrochemical feature importance matrix, an adaptive threshold method is used to select the optimal feature subset; The method of selecting the optimal feature subset based on the electrochemical feature importance matrix by using an adaptive threshold method includes: Design an electrochemical feature selection adaptive threshold function and calculate the electrochemical feature selection adaptive threshold; The electrochemical feature importance matrix is ​​traversed, and the electrochemical features greater than the electrochemical feature selection adaptive threshold are selected into the optimal feature subset.

[0090] Specifically, step S1313 performs feature selection on the multidimensional electrochemical feature matrix, the purpose of which is to screen out the most representative and discriminative feature subsets from the high-dimensional feature space, reduce feature redundancy, and improve the generalization performance of the subsequent health assessment model. Feature importance is an indicator that measures the contribution of a single feature to the prediction target. Common calculation methods include feature importance based on tree models (such as average impurity reduction), feature importance based on regularization (such as L1 regularization coefficient), etc. Considering that there may be complex nonlinear relationships and interactions between electrochemical features, the present invention adopts a feature importance measurement method based on Shapley value. The Shapley value originates from game theory and is used to evaluate the contribution of each participant in the game. By introducing it into the feature importance calculation, the average marginal contribution of each feature in all possible feature combinations can be measured, and the interaction effect between features is fully considered. Specifically, for each training sample, all possible feature subsets are calculated, the prediction performance of each subset is evaluated, and then the Shapley value calculation formula is used to quantify the average marginal contribution of each feature as its importance score. By executing this process on all training samples and taking the average, a robust feature importance measure can be obtained. Compared with traditional methods, the Shapley value can reveal the "net" contribution of the feature, overcome the influence of multicollinearity between features, and make the importance assessment more accurate and fair. The Shapley importance score of each feature is arranged according to the layout of the original feature matrix, that is, an importance matrix with the same shape as the multidimensional electrochemical feature matrix is ​​formed. This matrix intuitively depicts the relative contribution of different electrochemical features to health assessment and provides a quantitative basis for subsequent feature screening. By visually analyzing the importance matrix, such as drawing importance distribution diagrams and cluster heat maps, we can gain insight into the overall distribution characteristics of feature importance, discover possible feature clusters and outliers, and provide prior knowledge for the formulation of feature selection strategies.

[0091] The key to feature selection is to determine the optimal feature combination so that it can minimize redundancy and noise while retaining the most original information. Traditional feature selection methods are usually based on preset fixed thresholds, such as variance thresholds, correlation coefficient thresholds, etc., which lack flexibility and adaptability. Considering the diversity and non-stationarity of the distribution of electrochemical feature importance, the present invention designs an adaptive threshold feature selection method. Specifically, an adaptive threshold function threshold1= base1 is defined. value + k × range, where base valueis the basic threshold, range is the value range of the importance score, and k is the adaptive coefficient. By optimizing the k value, the threshold can be adaptively adjusted to the dynamic distribution of the importance score, avoiding the selection of too many redundant features while retaining key information features. After determining the optimal k value, the electrochemical feature importance matrix is ​​traversed, and all features with importance scores greater than the adaptive threshold are selected to form the final feature subset. This method can automatically adjust the granularity of feature selection from a data-driven perspective, overcome the subjectivity and limitations of manually setting thresholds, and make the selection results more robust and reliable.

[0092] Through the above steps, the automated screening from the high-dimensional electrochemical feature matrix to the low-dimensional optimal feature subset is realized, which effectively reduces the redundancy of the feature space and highlights the most decisive key information for health assessment. Compared with the traditional feature selection method, the innovation of the present invention is embodied in: (1) The Shapley value is used to measure the feature importance, which overcomes the influence of multicollinearity between features and obtains more accurate and fair importance assessment results; (2) The electrochemical feature importance matrix is ​​constructed to intuitively characterize the characteristics of the importance distribution and provide prior knowledge for threshold selection; (3) The adaptive threshold method is introduced to automatically adjust the feature selection granularity according to the dynamic characteristics of the importance distribution, and obtain a more robust and reliable feature subset. These innovations are organically combined to form a set of data-driven electrochemical feature selection methods, which lays a high-quality data foundation for the subsequent construction of health assessment models.

[0093] Example 3 This embodiment provides a multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells on the basis of Embodiment 2, including: Step S1313, performing feature selection on the multi-dimensional electrochemical feature matrix; Further, step S1313 includes: Step S13131, calculating the importance score of each electrochemical feature in the multidimensional electrochemical feature matrix; Step S13132, constructing an electrochemical feature importance matrix based on the importance score of each electrochemical feature; Step S13133, based on the electrochemical feature importance matrix, an adaptive threshold method is used to select the optimal feature subset; The method of selecting the optimal feature subset based on the electrochemical feature importance matrix by using an adaptive threshold method includes: Design an electrochemical feature selection adaptive threshold function and calculate the electrochemical feature selection adaptive threshold; The electrochemical feature importance matrix is ​​traversed, and the electrochemical features greater than the electrochemical feature selection adaptive threshold are selected into the optimal feature subset.

[0094] The electrochemical feature selection adaptive threshold function includes: Based on the importance scores of the electrochemical features, the skewness coefficient was calculated; Calculate the interquartile range of all electrochemical feature importance scores ,according to , get the dispersion coefficient; Obtain the number of columns of the electrochemical feature importance matrix, and calculate the logarithm of the number of features according to the number of columns of the electrochemical feature importance matrix; The adaptive threshold for electrochemical feature selection is obtained by combining the skewness coefficient, dispersion coefficient and logarithmically weighted number of features.

[0095] ; in: Threshold: Adaptive threshold for electrochemical feature selection.

[0096] : Importance score of the ith electrochemical feature.

[0097] : The mean of all electrochemical feature importance scores.

[0098] n: The total number of electrochemical features, i.e., the number of columns in the electrochemical feature importance matrix.

[0099] : Standard deviation of all electrochemical feature importance scores.

[0100] : The interquartile range of all electrochemical feature importance scores, obtained by taking the difference between the first and third quartiles of the electrochemical feature importance matrix.

[0101] : Skewness coefficient, which reflects the asymmetry of the importance score distribution; : The weight of the skewness coefficient controls the influence of the skewness coefficient on the threshold. The larger the value, the more attention is paid to the asymmetry of the importance score distribution, and it is hoped that important features can be retained more selectively by increasing the threshold.

[0102] : Dispersion coefficient, which reflects the degree of dispersion of the importance score distribution.

[0103] : The weight of the discrete coefficient controls the influence of the discrete coefficient on the threshold. The larger the value, the more attention is paid to the discreteness of the importance score, and it is hoped that important features can be retained more comprehensively by lowering the threshold.

[0104] : The logarithm of the number of features, reflecting the complexity of the feature selection task; : The weight of the logarithm of the number of features, which controls the impact of the number of features on the threshold. The larger the value, the more attention is paid to the complexity of the feature selection task, and the size of the selected feature subset is controlled by increasing the threshold.

[0105] s: The scale parameter in the dispersion coefficient controls the sensitivity of the ratio of the interquartile range to the standard deviation to the dispersion coefficient. The larger the s value, the less sensitive the dispersion coefficient is to changes in the ratio of the interquartile range to the standard deviation, and the more adaptable the threshold function is to the degree of dispersion of the importance score.

[0106] These four hyperparameters together determine the adaptability and selection strategy of the threshold function to the electrochemical feature importance matrix. By adjusting these hyperparameters, the threshold function can achieve the best feature selection performance under different data sets and task requirements. Hyperparameters can be tuned by methods such as cross-validation.

[0107] The formula measures the asymmetry of the importance score distribution by calculating the skewness coefficient of the importance score (i.e., the third center distance divided by the cube of the standard deviation). The larger the skewness coefficient, the more asymmetric the distribution is, and the threshold should be adjusted accordingly to more selectively retain important features. The discreteness of the importance score distribution is measured by calculating the negative exponent of the ratio of the interquartile range of the importance score to the standard deviation. The greater the degree of dispersion, the more obvious the difference between important and unimportant features. The complexity of the feature selection task is measured by taking the logarithm of the total number of features. The more features there are, the higher the complexity of the task, and the threshold should be increased accordingly to ensure that the selected features are important enough while controlling the size of the subset. Through the weighted combination of the above three factors, the adaptive threshold function can dynamically adjust the threshold according to the actual distribution characteristics of the electrochemical feature importance score, so as to achieve more flexible and effective feature selection.

[0108] The adaptive threshold function for electrochemical feature selection can automatically adjust the threshold according to the statistical characteristics of the importance matrix of different electrochemical features, without manually setting a fixed threshold, thus improving the adaptive ability of feature selection. By introducing the higher-order moments (skewness) and dispersion measures (interquartile range) of the importance score distribution, the threshold function is more robust to outliers and noise, and can make reliable selection decisions even when the importance score distribution is uneven or there are outliers. The design of the threshold function takes into account multiple key features of the importance score distribution, such as skewness, dispersion, number of features, etc. These features have clear statistical significance, making the behavior of the threshold function more transparent and explainable.

[0109] As for the output of the threshold function, as the skewness of the importance score distribution increases, the discreteness decreases, and the number of features increases, the output value of the threshold function will increase accordingly. This means that when the difference between important and unimportant features is less obvious and the number of features is greater, the threshold function will automatically increase the selection criteria to ensure that the selected feature subset is sufficiently refined and important. Conversely, when the difference between important and unimportant features is more obvious and the number of features is smaller, the threshold function will appropriately lower the selection criteria to avoid missing valuable features. In short, the electrochemical feature selection adaptive threshold function can dynamically adjust the selection threshold according to the actual distribution characteristics of the electrochemical feature importance matrix, adaptively balance the size and importance of the feature subset, improve the flexibility, robustness and interpretability of feature selection, and provide a better feature subset for the subsequent construction of the battery health assessment model.

[0110] Example 4 This embodiment provides a multi-dimensional parameter evaluation system based on thermal runaway risk detection of battery cells on the basis of embodiment 1, such as Fig. 9 As shown, including: Feature space construction module: used to synchronously collect multi-dimensional parameter data of battery cells under different charge and discharge states to construct a thermal-electric coupling data set; obtain electrochemical impedance spectroscopy data of battery cells under different aging states, and construct a refined multi-dimensional electrochemical feature space based on the thermal-electric coupling data set and electrochemical impedance spectroscopy data; Mapping relationship establishment module: Based on the refined multi-dimensional electrochemical feature space, build and train the battery cell health assessment model and extract the battery cell health indicators; establish the battery cell thermal runaway risk level and construct the mapping relationship between the battery cell health indicators and the battery cell thermal runaway risk level; Thermal runaway judgment module: obtains the real-time operating parameters of the battery cell, and obtains the operating condition adaptive battery cell health index according to the real-time operating condition parameters of the battery cell and the battery cell health assessment model; obtains the predicted thermal runaway risk level according to the operating condition adaptive battery cell health index and the mapping relationship between the battery cell health index and the thermal runaway risk level; determines whether the predicted thermal runaway risk level exceeds the preset threshold, and if so, triggers the early warning mechanism.

[0111] In the feature space construction module, the multidimensional parameter data includes multi-point temperature data, load current characteristic data and potential response data; the construction of a refined multidimensional electrochemical feature space based on the thermal-electric coupling data set and the electrochemical impedance spectroscopy data includes: Step S1210, obtaining a nonlinear frequency response characteristic matrix based on the thermal-electric coupling data set; Step S1220, obtaining electrochemical impedance spectroscopy data of the battery cell under different aging conditions; Step S1230, extracting key impedance parameters from electrochemical impedance spectroscopy data and constructing an EIS characteristic matrix; Step S1310, integrating the nonlinear frequency response characteristic matrix and the EIS characteristic matrix to construct a refined multi-dimensional electrochemical characteristic space.

[0112] The step S1210 includes: Step S1211, performing time-frequency analysis on the thermal-electric coupling data set to extract multi-parameter wavelet coefficients; the multi-parameter wavelet coefficients include temperature wavelet coefficients, current wavelet coefficients, and voltage wavelet coefficients; Step S1212: obtaining a nonlinear frequency response characteristic matrix based on the multi-parameter wavelet coefficients.

[0113] The step S1211 includes: Step S12111, performing time-frequency analysis on the multi-point temperature data to extract temperature wavelet coefficients; Step S12112, performing time-frequency analysis on the load current characteristic data to extract current wavelet coefficients; Step S12113, perform time-frequency analysis on the potential response data and extract voltage wavelet coefficients.

[0114] The step S1212 includes: Step S12121, based on the temperature wavelet coefficient, calculate the energy distribution of the multi-point temperature data in different frequency bands to obtain the temperature frequency domain feature vector; Step S12122, based on the current wavelet coefficient, calculate the energy distribution of the load current characteristic data in different frequency bands to obtain the current frequency domain characteristic vector; Step S12123, based on the voltage wavelet coefficients, calculating the energy distribution of the potential response data in different frequency bands to obtain the voltage frequency domain feature vector; Step S12124, concatenate the temperature frequency domain characteristic vector, the current frequency domain characteristic vector and the voltage frequency domain characteristic vector by column to form a nonlinear frequency response characteristic matrix.

[0115] The step S1310 includes: Step S1311, performing data standardization processing on the nonlinear frequency response characteristic matrix and the EIS characteristic matrix; Step S1312, splicing the standardized nonlinear frequency response characteristic matrix and the EIS characteristic matrix by column to form a multidimensional electrochemical characteristic matrix; each row of the multidimensional electrochemical characteristic matrix corresponds to a battery cell sample, and each column corresponds to an electrochemical characteristic; Step S1313, performing feature selection on the multi-dimensional electrochemical feature matrix; Step S1314, performing feature extraction on the multidimensional electrochemical feature matrix after feature selection to obtain a refined multidimensional electrochemical feature space, wherein the refined multidimensional electrochemical feature space contains n1 electrochemical features.

[0116] In the mapping relationship establishment module, the construction and training of the battery cell health assessment model includes: Step S1321, constructing a training sample set using the electrochemical features in the refined multi-dimensional electrochemical feature space as sample features and the manually labeled cell health indicators as sample labels; Step S1322, using a weighted linear regression model to construct an initial cell health assessment model; Step S1323, using the training sample set to train the initial cell health assessment model to obtain the weight coefficient and bias term b of each electrochemical feature; Step S1324, based on the weight coefficient of each electrochemical feature and the bias term b, a final cell health assessment model is obtained.

[0117] In the mapping relationship establishment module, the mapping relationship between the battery cell health index and the battery cell thermal runaway risk level is constructed including: Step S1331, obtaining the material characteristics and design parameters of the battery cell, analyzing the thermal runaway mechanism of the battery cell, and determining the key factors affecting the thermal runaway; Step S1332, classifying the thermal runaway risk level; Step S1333: Based on the key factors affecting thermal runaway and the risk level classification, a mapping relationship between the battery cell health index and the thermal runaway risk level is established.

[0118] In the thermal runaway judgment module, the condition-adaptive battery health index is obtained according to the real-time battery condition parameters and the battery health evaluation model, including: Step S2110, collecting real-time operating parameters of the battery cell, wherein the real-time operating parameters of the battery cell include ambient temperature, charge and discharge rate, and battery cell SOC; Step S2120, dynamically adjust the weight coefficient of each electrochemical characteristic in the battery cell health assessment model in combination with the real-time operating condition parameters of the battery cell to obtain an operating condition adaptive battery cell health index.

[0119] The step S2120 includes: Step S2121, obtaining the operating parameters of the battery cell, and constructing a mapping relationship between the operating parameters and the weight adjustment coefficient; Step S2122, calculating the real-time weight adjustment coefficient according to the real-time operating parameters of the battery cell and the mapping relationship between the operating parameters and the weight adjustment coefficient; Step S2123, adjusting the weight coefficient of each electrochemical feature in the battery cell health assessment model according to the real-time weight adjustment coefficient to obtain an adjusted weight coefficient; Step S2124, using the adjusted weight coefficient, calculate the operating condition adaptive battery cell health index under the current operating condition.

[0120] The method and system of the present application may be implemented in many ways. For example, the method and system of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above, unless otherwise specifically stated.

[0121] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0122] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells, characterized in that: The method comprises: Synchronously collect multi-dimensional parameter data of battery cells under different charge and discharge states to construct a thermal-electric coupling data set; obtain electrochemical impedance spectroscopy data of battery cells under different aging states, and construct a refined multi-dimensional electrochemical feature space based on the thermal-electric coupling data set and electrochemical impedance spectroscopy data; construct and train a battery cell health assessment model based on the refined multi-dimensional electrochemical feature space, and extract battery cell health indicators; establish a battery cell thermal runaway risk level, and construct a mapping relationship between battery cell health indicators and battery cell thermal runaway risk level; The real-time operating condition parameters of the battery cell are obtained, and the operating condition adaptive battery cell health index is obtained according to the real-time operating condition parameters of the battery cell and the battery cell health assessment model; the predicted thermal runaway risk level is obtained according to the operating condition adaptive battery cell health index and the mapping relationship between the battery cell health index and the thermal runaway risk level; it is determined whether the predicted thermal runaway risk level exceeds a preset threshold, and if so, a warning mechanism is triggered.

2. The multidimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 1, characterized in that: The multi-dimensional parameter data includes multi-point temperature data, load current characteristic data and potential response data; The method of constructing a refined multi-dimensional electrochemical feature space based on the thermal-electric coupling data set and the electrochemical impedance spectroscopy data includes: Based on the thermal-electric coupling data set, the nonlinear frequency response characteristic matrix is ​​obtained; Based on the electrochemical impedance spectroscopy data, the EIS characteristic matrix is ​​obtained; The nonlinear frequency response characteristic matrix and the EIS characteristic matrix are integrated to construct a refined multi-dimensional electrochemical characteristic space.

3. The multidimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 2 is characterized in that: The nonlinear frequency response characteristic matrix obtained based on the thermal-electric coupling data set includes: Performing time-frequency analysis on the thermal-electric coupling data set to extract multi-parameter wavelet coefficients; the multi-parameter wavelet coefficients include temperature wavelet coefficients, current wavelet coefficients, and voltage wavelet coefficients; Based on the multi-parameter wavelet coefficients, the nonlinear frequency response characteristic matrix is ​​obtained.

4. The multidimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 3 is characterized in that: The extracting of multi-parameter wavelet coefficients comprises: Perform time-frequency analysis on multi-point temperature data and extract temperature wavelet coefficients; Perform time-frequency analysis on load current characteristic data and extract current wavelet coefficients; The potential response data is subjected to time-frequency analysis and the voltage wavelet coefficients are extracted.

5. The multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 3 is characterized in that: The method of obtaining a nonlinear frequency response characteristic matrix based on multi-parameter wavelet coefficients includes: Based on the temperature wavelet coefficient, the energy distribution of multi-point temperature data in different frequency bands is calculated to obtain the temperature frequency domain feature vector; Based on the current wavelet coefficient, the energy distribution of the load current characteristic data in different frequency bands is calculated to obtain the current frequency domain characteristic vector; Based on the voltage wavelet coefficient, the energy distribution of the potential response data in different frequency bands is calculated to obtain the voltage frequency domain feature vector; The temperature frequency domain eigenvector, the current frequency domain eigenvector and the voltage frequency domain eigenvector are concatenated column by column to form a nonlinear frequency response characteristic matrix.

6. The multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 2, characterized in that: The EIS characteristic matrix obtained based on the electrochemical impedance spectroscopy data includes: Perform equivalent circuit fitting on electrochemical impedance spectroscopy data to obtain equivalent circuit model parameters; Extract key impedance parameters from equivalent circuit model parameters; The EIS characteristic matrix is ​​constructed from key impedance parameters.

7. The multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 2, characterized in that: The method of fusing the nonlinear frequency response characteristic matrix and the EIS characteristic matrix to construct a refined multi-dimensional electrochemical characteristic space includes: Perform data standardization on the nonlinear frequency response characteristic matrix and the EIS characteristic matrix; The standardized nonlinear frequency response characteristic matrix and the EIS characteristic matrix are spliced ​​by column to form a multidimensional electrochemical characteristic matrix, wherein each row of the multidimensional electrochemical characteristic matrix corresponds to a battery cell sample, and each column corresponds to an electrochemical characteristic; Perform feature selection on multidimensional electrochemical feature matrix; Feature extraction is performed on the multidimensional electrochemical feature matrix after feature selection to obtain a refined multidimensional electrochemical feature space, wherein the refined multidimensional electrochemical feature space contains n1 electrochemical features.

8. The multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 7, characterized in that: The feature selection of the multidimensional electrochemical feature matrix comprises: Calculate the importance score of each electrochemical feature in the multidimensional electrochemical feature matrix; Based on the importance score of each electrochemical feature, an electrochemical feature importance matrix is ​​constructed; Based on the electrochemical feature importance matrix, an adaptive threshold method was used to select the optimal feature subset.

9. The multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 8, characterized in that: The method of selecting the optimal feature subset based on the electrochemical feature importance matrix by using an adaptive threshold method includes: Design an electrochemical feature selection adaptive threshold function and calculate the electrochemical feature selection adaptive threshold; The electrochemical feature importance matrix is ​​traversed, and the electrochemical features greater than the electrochemical feature selection adaptive threshold are selected into the optimal feature subset.

10. The multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 7, characterized in that: The construction and training of the battery cell health assessment model based on the refined multi-dimensional electrochemical feature space includes: The electrochemical features in the refined multi-dimensional electrochemical feature space are used as sample features, and the manually labeled battery cell health indicators are used as sample labels to construct a training sample set; A weighted linear regression model is used to build an initial battery cell health assessment model; Using the training sample set, train the initial cell health assessment model to obtain the weight coefficient and bias term b of each electrochemical feature; Based on the weight coefficient and bias term b of each electrochemical feature, the final cell health assessment model is obtained.

11. The multi-dimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to claim 10, characterized in that: The condition-adaptive battery health index is obtained according to the battery real-time condition parameters and the battery health evaluation model, including: Obtain the operating parameters of the battery cell and construct a mapping relationship between the operating parameters and the weight adjustment coefficient; Calculate the real-time weight adjustment coefficient according to the real-time operating parameters of the battery cell and the mapping relationship between the operating parameters and the weight adjustment coefficient; According to the real-time weight adjustment coefficient, the weight coefficient of each electrochemical feature in the battery cell health assessment model is adjusted to obtain an adjusted weight coefficient; Use the adjusted weight coefficient to calculate the condition-adaptive cell health index under the current condition.

12. A multidimensional parameter evaluation system based on thermal runaway risk detection of battery cells, which is used to implement the multidimensional parameter evaluation method based on thermal runaway risk detection of battery cells according to any one of claims 1 to 11, characterized in that: The system comprises: Feature space construction module: used to synchronously collect multi-dimensional parameter data of battery cells under different charge and discharge states to construct a thermal-electric coupling data set; obtain electrochemical impedance spectroscopy data of battery cells under different aging states, and construct a refined multi-dimensional electrochemical feature space based on the thermal-electric coupling data set and electrochemical impedance spectroscopy data; Mapping relationship establishment module: Based on the refined multi-dimensional electrochemical feature space, build and train the battery cell health assessment model and extract the battery cell health indicators; establish the battery cell thermal runaway risk level and construct the mapping relationship between the battery cell health indicators and the battery cell thermal runaway risk level; Thermal runaway judgment module: obtains the real-time operating parameters of the battery cell, and obtains the operating condition adaptive battery cell health index according to the real-time operating condition parameters of the battery cell and the battery cell health assessment model; obtains the predicted thermal runaway risk level according to the operating condition adaptive battery cell health index and the mapping relationship between the battery cell health index and the thermal runaway risk level; determines whether the predicted thermal runaway risk level exceeds the preset threshold, and if so, triggers the early warning mechanism.

Images