A power battery thermal runaway early warning method and system based on singular spectrum entropy

Abstract

The application provides a power battery thermal runaway early warning method and system based on singular spectrum entropy, the method collects and uploads output signals of each sensor of the power battery from a vehicle end, constructs a mode matrix according to a time domain signal sequence formed in a time sequence of signal transmission, performs singular value decomposition on the mode matrix, arranges singular values to form a singular spectrum of the signal, performs singular spectrum analysis on the signal, and then calculates a singular spectrum entropy of the time domain signal sequence, performs normalization processing on the singular spectrum entropy, quantitatively evaluates an alarm coefficient, compares the alarm coefficient with an alarm threshold, and thereby judges a thermal runaway risk of the power battery and locates a fault position. The application analyzes from the information amount contained in the signal, uses the singular spectrum entropy value to early warn the power battery thermal runaway, greatly improves the sensitivity to the fault signal, can more quickly detect the time when the risk occurs and locate the fault monomer, and effectively advances the time point of the thermal runaway early warning.

Description

Technical Field

[0001] This invention relates to the field of power battery technology, specifically to a method and system for early warning of thermal runaway in power batteries based on singular spectral entropy. Background Technology

[0002] To alleviate the increasingly serious environmental problems and the scarcity of non-renewable energy, the electrification of automotive power systems has become one of the main development directions for future automobiles. Power batteries, as the energy storage device and sole power source of electric vehicles, occupy a core position in their development. However, safety incidents involving power batteries frequently occur during their use, typically manifesting as thermal runaway. This runaway is mainly characterized by a sudden increase in cell temperature, leading to fire, gas leaks, and explosions, which can easily cause personal injury and property damage.

[0003] Existing methods for early warning of thermal runaway risk are mainly based on temperature thresholds or temperature change rate thresholds. When the temperature or temperature change rate of the power battery exceeds the set threshold, an alarm signal is issued, thus triggering a thermal runaway warning. However, when these limits are exceeded, thermal runaway is almost inevitable, and the losses are difficult to recover. Therefore, such solutions have significant limitations. In addition, traditional signal feature extraction methods have the following shortcomings: (1) they are difficult to accurately describe the nonlinear characteristics of the system and the non-stationary signal characteristics; (2) they are difficult to solve the problem of quantitative evaluation of signal characteristics. Therefore, a power battery thermal runaway early warning scheme that can accurately and quantitatively evaluate the thermal runaway characteristics of the signal and provide more timely warnings is urgently needed. Summary of the Invention

[0004] To address the shortcomings of current power battery thermal runaway early warning systems, such as insufficient timeliness, difficulty in accurately describing nonlinear systems, and challenges in quantitatively evaluating signal characteristics, this invention proposes a power battery thermal runaway early warning method based on singular spectral entropy. This method utilizes the properties of singular spectral entropy to detect thermal runaway fault information from the perspective of information quantity, based on information collected by sensors. It can quickly detect and locate the current thermal runaway faulty cell, allowing for earlier prediction of battery thermal runaway. This invention also relates to a power battery thermal runaway early warning system based on singular spectral entropy.

[0005] The technical solution of the present invention is as follows:

[0006] A method for early warning of thermal runaway in power batteries based on singular spectral entropy, characterized by comprising the following steps:

[0007] The process of constructing the pattern matrix involves collecting and uploading the output signals of each sensor in the power battery from the vehicle, and constructing the pattern matrix according to the time-domain signal sequence formed by the time sequence of the signals.

[0008] The singular value decomposition step involves performing singular value decomposition on the mode matrix, arranging the singular values ​​to form the singular spectrum of the signal, and the number of non-zero singular values ​​is equal to the total number of modes contained in each column of the mode matrix.

[0009] The steps for calculating singular spectral entropy involve performing singular spectral analysis on the signal and calculating the proportion of each mode in all modes, thereby calculating the singular spectral entropy of the time-domain signal sequence to reflect the uncertainty of each mode under the singular spectral division of the time-domain signal sequence and to quantitatively describe the complex temporal morphological characteristics of the signal energy distribution.

[0010] The normalization processing early warning analysis step involves normalizing the singular spectral entropy of the obtained time-domain signal sequence based on a comparison with the singular spectral entropy of white noise. The alarm coefficient is then quantitatively evaluated based on the normalized singular spectral entropy and compared with the alarm threshold to determine the risk of thermal runaway of the power battery and locate the fault location.

[0011] Preferably, in the step of constructing the pattern matrix, the pattern matrix is ​​constructed by setting the analysis window length and the time delay constant and extracting the time domain signal sequence in window order.

[0012] Preferably, in the step of calculating the singular spectral entropy, the information entropy of the singular spectrum is calculated based on the proportion of each mode in all modes, according to the definition of Shannon entropy, so as to obtain the singular spectral entropy of the time-domain signal sequence.

[0013] Preferably, in the normalization processing early warning analysis step, the Z-score method is used to standardize the normalized singular spectral entropy. The difference between the normalized singular spectral entropy of a certain power battery cell and the average value of the normalized singular spectral entropy of all power battery cells is divided by the standard deviation of the normalized singular spectral entropy of all power battery cells to obtain the alarm coefficient of each power battery cell. The alarm coefficient of each power battery cell is compared with the alarm threshold to determine the thermal runaway risk of the power battery and locate the fault location.

[0014] Preferably, in the step of constructing the mode matrix, the output signals of each sensor of the power battery include voltage, current, and temperature signals under several actual operating conditions or charging conditions when the vehicle is in motion.

[0015] Preferably, in the normalization processing early warning analysis step, the driving data of the power battery cloud big data platform is analyzed, verified and adjusted to determine the alarm threshold range.

[0016] A power battery thermal runaway early warning system based on singular spectral entropy is characterized by comprising a mode matrix construction module, a singular value decomposition module, a singular spectral entropy calculation module, and a normalization processing early warning analysis module connected in sequence.

[0017] The mode matrix construction module collects and uploads the output signals of each sensor of the power battery from the vehicle and constructs the mode matrix according to the time domain signal sequence formed by the time order of signal input.

[0018] The singular value decomposition module performs singular value decomposition on the mode matrix, arranges the singular values ​​to form the singular spectrum of the signal, and the number of non-zero singular values ​​is the total number of modes contained in each column of the mode matrix.

[0019] The module for calculating singular spectral entropy performs singular spectral analysis of the signal and calculates the proportion of each mode in all modes, thereby calculating the singular spectral entropy of the time-domain signal sequence to reflect the uncertainty of each mode under the singular spectral division of the time-domain signal sequence and to quantitatively describe the complex temporal morphological characteristics of the signal energy distribution.

[0020] The normalization processing early warning analysis module normalizes the singular spectral entropy of the obtained time-domain signal sequence based on a comparison with the singular spectral entropy of white noise. It then quantitatively evaluates the alarm coefficient based on the normalized singular spectral entropy and compares it with the alarm threshold to determine the risk of thermal runaway of the power battery and locate the fault position.

[0021] Preferably, the pattern matrix construction module constructs the pattern matrix by setting the analysis window length and the time delay constant and extracting the time domain signal sequence in window order.

[0022] Preferably, the singular spectrum entropy calculation module calculates the information entropy of the singular spectrum based on the proportion of each mode in all modes, according to the definition of Shannon entropy, thereby obtaining the singular spectrum entropy of the time-domain signal sequence.

[0023] Preferably, the normalization processing early warning analysis module uses the Z-score method to standardize the normalized singular spectrum entropy. The difference between the normalized singular spectrum entropy of a certain power battery cell and the average of the normalized singular spectrum entropies of all power battery cells is divided by the standard deviation of the normalized singular spectrum entropies of all power battery cells to obtain the alarm coefficient of each power battery cell. The alarm coefficient of each power battery cell is compared with the alarm threshold to determine the thermal runaway risk of the power battery and locate the fault location.

[0024] The beneficial effects of this invention are as follows:

[0025] This invention proposes a method for early warning of thermal runaway in power batteries based on singular spectral entropy. This method combines singular spectrum and information entropy, using the singular value decomposition result as the object of information entropy calculation, specifically calculating the information entropy of the singular spectrum (referred to as singular spectral entropy). It is particularly suitable for applications with a small number of sampling points and where the acquired signals contain noise. Singular spectral analysis of signals is a modern spectral analysis technique based on dynamic analysis. Its basic idea is to obtain the inherent complexity characteristics of the time-domain signal sequence of a power battery system by performing phase space reconstruction and singular value decomposition. Based on singular spectral analysis, calculating singular spectral entropy can quantitatively describe the complex state characteristics of the time series. In information theory, information entropy describes the degree of uncertainty of a system, used to measure the changes in a signal and the amount of information it contains. Singular spectral entropy is essentially a measure of information content; it reflects the degree of uncertainty of each mode in the time-domain signal sequence under singular spectral division, and also embodies the temporal complexity of the signal energy distribution. Under normal operating conditions, the signal complexity is low, and the energy is concentrated in a few modes, resulting in a small singular spectral entropy value. Conversely, when a battery cell experiences thermal runaway, the complexity of the incoming signal increases, and the energy of complex transient signals is more dispersed, leading to a larger singular spectral entropy value. Furthermore, the singular spectral entropy value is highly sensitive to abnormal changes in the input signal, allowing for rapid detection and location of the faulty cell. Compared to traditional algorithms that directly set thresholds to determine whether voltage, temperature, and other data exceed thresholds, this significantly advances the thermal runaway alarm, providing valuable time for fault handling and greatly improving safety. Therefore, this invention utilizes the characteristic that singular spectral entropy can quickly reflect changes in the internal information of the power battery system and is independent of signal stability, enabling rapid determination of whether the current battery operating state is abnormal.

[0026] This invention constructs a pattern matrix by combining information collected and uploaded by sensors with a phase space reconstruction step. This is achieved by constructing a pattern matrix based on the time-domain signal sequence formed by the sequential input of signals. Singular value decomposition (SVD) is then performed to construct the singular spectrum of the signal. Based on the SVD analysis, the singular spectrum entropy is calculated. From the perspective of information content, the properties of information entropy are used to detect fault information, describing runaway information with an information content measure. This allows for early identification and warning of thermal runaway based on information content measurement. Vehicle-side information acquisition inevitably involves noise interference, and due to performance limitations, the sampling frequency is lower than that under laboratory conditions. Singular spectrum entropy is well-suited for applications with fewer sampling points and noisy signals. The value of the singular spectrum entropy changes significantly when abnormal information is present, allowing for rapid determination of whether the current battery operating state is abnormal. The normalization processing early warning analysis step normalizes the singular spectral entropy of the obtained time-domain signal sequence by comparing it with the singular spectral entropy of white noise. The alarm coefficient is then quantitatively evaluated based on the normalized singular spectral entropy. This invention defines an alarm coefficient as a parameter based on the singular spectral entropy value. The alarm coefficient value characterizes the intrinsic information of the power battery and can quantitatively describe the complex state characteristics of the time series. Therefore, the alarm coefficient parameter can better quantify the risk of thermal runaway. Then, by comparing the alarm coefficient value of each power battery cell with a preset alarm threshold, it is determined whether the battery cell has experienced thermal runaway and its location is determined. This allows for rapid and accurate identification and quantitative analysis of the power battery experiencing thermal runaway. When a thermal runaway risk is detected at a certain moment, this scheme can quickly locate the power battery cell experiencing the anomaly at that moment, greatly improving the system's sensitivity to fault signals. It can more quickly detect the moment of risk occurrence and locate the faulty cell, effectively advancing the thermal runaway warning time, significantly improving safety, and fundamentally solving the problems of accurately describing nonlinear systems and quantitatively evaluating signal characteristics. It has good application prospects.

[0027] Preferably, this invention constructs a pattern matrix by setting the analysis window length and time delay constant and extracting time-domain signal sequences in window order. When constructing the pattern matrix, the value of the time delay constant ensures that no information is missing from the signal, fully reflecting the characteristics of the data. After calculating the singular spectral entropy value, normalization is performed based on a comparison with the singular spectral entropy of white noise. This not only facilitates a direct comparison of signal complexity but also eliminates the influence of the analysis window length selection on the calculation results. In the normalization processing early warning analysis step, when quantitatively evaluating the alarm coefficient using singular spectral entropy and comparing it with the alarm threshold to determine the risk of thermal runaway of the power battery, the selection of the alarm threshold can be determined based on the analysis and verification of a large amount of driving data from the power battery cloud big data platform, thereby expanding the scope of application and reducing the false alarm and missed alarm rates.

[0028] This invention proposes a power battery thermal runaway early warning system based on singular spectral entropy. Corresponding to the aforementioned power battery thermal runaway early warning method based on singular spectral entropy, this system can be understood as an implementation system of the singular spectral entropy-based power battery thermal runaway early warning method. It includes a module for constructing a pattern matrix, a singular value decomposition module, a module for calculating singular spectral entropy, and a normalization processing and early warning analysis module, connected sequentially. These modules work collaboratively to determine the thermal runaway of the power battery. By extracting signals transmitted from sensors and performing phase space reconstruction and singular value decomposition, abnormal signals are identified and analyzed. Based on the singular spectral analysis of the signals, singular spectral entropy is calculated, analyzing the information content of the signals. By comparing the difference in information content between fault signals and normal signals, thermal runaway is warned. This scheme significantly improves the system's sensitivity to fault signals, enabling faster detection of the risk occurrence and location of the faulty individual cell, effectively advancing the thermal runaway early warning time and greatly improving safety. Attached Figure Description

[0029] Figure 1 This is a flowchart of the power battery thermal runaway early warning method based on singular spectral entropy according to the present invention.

[0030] Figure 2 This is a flowchart of the singular spectral entropy optimization calculation process in the power battery thermal runaway early warning method based on singular spectral entropy of the present invention.

[0031] Figure 3 This is a voltage distribution curve of all cells in a certain group of power batteries at the time of a fault, according to the present invention.

[0032] Figure 4 This is a graph showing the singular spectral entropy distribution of all cells in a certain group of power batteries at the moment of failure.

[0033] Figure 5 This is a graph showing the change of the characteristic value S of the faulty unit and the two normal units of the present invention over time. Detailed Implementation

[0034] The present invention will now be described with reference to the accompanying drawings.

[0035] A method for early warning of thermal runaway in power batteries based on singular spectral entropy, such as... Figure 1As shown, the process includes a step of constructing a mode matrix, where the output signals of each sensor in the power battery are collected and uploaded from the vehicle, i.e., the input signals of each sensor, and data preprocessing (voltage, temperature, current, etc.) can be performed. The mode matrix is ​​constructed according to the time-domain signal sequence formed by the time sequence of signal input. A singular value decomposition step is performed on the mode matrix to decompose the singular values, arranging the singular values ​​to form the singular spectrum of the signal, and the number of non-zero singular values ​​is equal to the total number of modes contained in each column of the mode matrix. A singular spectrum entropy calculation step is performed to analyze the singular spectrum of the signal and calculate the proportion of each mode in all modes, and then calculate the entropy. The method obtains the singular spectral entropy (i.e., calculates the entropy value) of the time-domain signal sequence to reflect the uncertainty of each mode under the singular spectrum division and to quantitatively describe the complex temporal morphological characteristics of the signal energy distribution. The normalization processing early warning analysis step normalizes the obtained singular spectral entropy of the time-domain signal sequence based on a comparison with the singular spectral entropy of white noise. The alarm coefficient is quantitatively evaluated based on the normalized singular spectral entropy and compared with an alarm threshold. If the entropy exceeds the alarm threshold, the faulty cell is located and a danger moment is alerted; otherwise, a diagnostic result is output, thereby determining the thermal runaway risk of the power battery and locating the fault location. This invention's power battery thermal runaway early warning method based on singular spectral entropy can be understood as employing the singular spectral entropy method. Simply put, the singular spectral entropy method is an analysis method that combines singular spectrum and information entropy, using the singular value decomposition result as the information entropy solution object. It is suitable for applications with a small number of sampling points and where the acquired signal contains noise. Singular spectrum analysis of signals is a modern spectral analysis technique based on dynamic analysis. Its basic idea is to obtain the inherent complexity characteristics of a system's time-domain signal sequence through phase space reconstruction and singular value decomposition. Based on singular spectrum analysis, the information entropy of the singular spectrum (or singular spectrum entropy) can be calculated to quantitatively describe the complex state characteristics of the time series. In the detection of thermal runaway in power batteries, signals transmitted by sensors are extracted and processed to identify and analyze abnormal signals. This approach analyzes the signal from the perspective of its information content, providing early warning of thermal runaway by comparing the difference in information content between fault and normal signals. The simpler the signal, the smaller the singular spectrum entropy value; complex transient signals have more dispersed energy, resulting in a larger singular spectrum entropy value. Furthermore, the singular spectrum entropy value is highly sensitive to abnormal changes in the input signal, and can quickly detect and locate the current faulty cell. This method greatly improves the system's sensitivity to fault signals, and can detect the moment of risk and locate the faulty cell more quickly. Compared with the traditional algorithm that directly sets thresholds to determine whether data such as voltage and temperature exceed the threshold, it can greatly advance the thermal runaway alarm of the battery, buy valuable time for fault handling, and greatly improve safety.

[0036] I. Steps for constructing the pattern matrix

[0037] First, the data is collected and uploaded from the vehicle, with an upload interval of 30 seconds, covering various actual operating conditions during vehicle operation, as well as various states such as charging. Taking voltage as an example, the normalized singular spectral entropy value of each power battery cell is calculated, such as... Figure 2 The flowchart shown shows X = {x} k}, k=1,2,...,N are the discrete variables of the original voltage sequence.

[0038] The voltage pattern matrix is ​​constructed according to the time sequence of signal input. Preferably, the pattern matrix is ​​constructed by setting the analysis window length and time delay constant, and extracting the time-domain signal sequence in a window order. The analysis window length is chosen as M, and the time delay constant as δ. Sample data is extracted in a window order of (M, δ), i.e., X = {x k The sequence is divided into λ segments, and these segments form the pattern matrix A:

[0039]

[0040] in, This indicates rounding up. The window length M should generally not exceed N / 3. Simultaneously, it is...

[0041] To ensure that no information is missing from the signal, the time delay constant is usually set to δ = 1.

[0042] II. Singular Value Decomposition Steps

[0043] Perform singular value decomposition on the pattern matrix A, and let the number of singular values ​​be M. Then, decompose the singular values ​​σ... i If (i = 1, 2, ..., M) are arranged in descending order, then σ1 ≥ σ2 ≥ ... ≥ σ M At this time σ i The singular value spectrum constitutes the measured signal. If σ i If the number of non-zero singular values ​​in pattern matrix A is j, then the value of j represents the total number of patterns contained in each column of pattern matrix A.

[0044] III. Steps for calculating singular spectral entropy

[0045] Singular spectrum analysis of the signal is performed, and the proportion of each mode in all modes is calculated. The singular spectral entropy of the time-domain signal sequence is then calculated. This is derived from the singular values ​​σ. i Based on the correspondence between the j patterns in the pattern matrix A, we can see that the singular value spectrum {σ i In fact, it is a division of the measured signal in the time domain. If we let...

[0046]

[0047] Then p i Let p represent the proportion of the i-th mode among all modes. iThis also represents the proportion of the i-th mode in the pattern matrix A among all modes. Therefore, according to the definition of information entropy (Shannon entropy), the singular spectral entropy of a signal in the time domain can be obtained as:

[0048]

[0049] Singular spectral entropy essentially reflects the degree of uncertainty of each mode in a time-domain signal sequence under singular spectrum partitioning. From a complexity analysis perspective, singular spectral entropy also reflects the temporal complexity of signal energy distribution. When applied to assessing the operating state of a single battery cell, under normal operating conditions, the signal complexity is low, and the energy is more concentrated in a few modes, resulting in a smaller singular spectral entropy value. Conversely, when a battery cell experiences thermal runaway, the complexity of the incoming signal increases, and the energy of complex transient signals is more dispersed, leading to a larger singular spectral entropy value.

[0050] IV. Normalization Processing and Early Warning Analysis Steps

[0051] According to information theory, when entropy reaches its maximum value, the system is in a steady state, and the energy distribution is most uniform at this point. The energy distribution differences between the various modes of a white noise signal are minimal, and its singular spectrum is essentially a straight line. Therefore, the singular spectrum entropy of a white noise signal is the largest, as given by the following equation:

[0052]

[0053] The singular spectral entropy H obtained from the step of calculating singular spectral entropy. s Based on the singular spectral entropy of white noise Normalization not only facilitates a direct comparison of signal complexity but also eliminates the influence of the chosen analysis window length on the calculation results. The normalized singular spectral entropy value... The calculation expression is:

[0054]

[0055] The alarm coefficient is quantitatively evaluated based on the normalized singular spectral entropy and compared with the alarm threshold to determine the risk of thermal runaway of the power battery and locate the fault location.

[0056] like Figure 3 The figure shows the voltage distribution curves of all cells in a certain group of batteries at the time of failure. The voltage change trend of this cell can be seen from the figure. When the battery experiences thermal runaway, it will have a very obvious voltage drop change compared to other normal cells. Figure 4 The invention illustrates the magnitude of the singular spectral entropy values ​​of each cell at the moment of failure, calculated based on voltage. This clearly demonstrates a significant surge in entropy value for the faulty battery cell at that moment. This invention leverages this characteristic of singular spectral entropy in describing power battery cells to provide early warning of thermal runaway in power batteries.

[0057] Furthermore, in the normalization processing early warning analysis step, the Z-score method is used to standardize the normalized singular spectral entropy. The difference between the normalized singular spectral entropy of a certain power battery cell and the average of the normalized singular spectral entropies of all power battery cells is divided by the standard deviation of the normalized singular spectral entropies of all power battery cells to obtain the alarm coefficient of each power battery cell. By setting the alarm coefficient for quantitative evaluation, the identification and early warning of thermal runaway events can be realized. Its expression is:

[0058]

[0059] Where S is the standardized feature value, i.e., the quantitative evaluation value of the alarm coefficient. H is the current entropy value. ave For all monomers in the current test point The mean, σ E This represents the standard deviation of the entropy of all individuals in the current test point.

[0060] Among all vehicles, faulty and normal vehicles follow a normal distribution, and thermal runaway is a low-probability event. Therefore, the 3σ principle is used to constrain it. The 3σ principle can be simply described as follows: under the assumption of a normal distribution, values ​​outside three times the mean (σ) have a very low probability of occurrence, and can therefore be considered outliers (i.e., |S| = 3 in the above formula). After extensive testing with real-world data, it was found that using 3σ results in overly sensitive detection and a high false alarm rate. Setting |S| ≥ 7.5 provides a more objective assessment of impending thermal runaway in the battery. Figure 5 The curves showing the characteristic value S of the faulty cell and two normal cells changing over time are shown. The data cutoff point is the alarm time of the traditional method, while the singular spectrum entropy value had already detected the anomaly before that. Clearly, compared to the traditional method, the method of this invention can detect the thermal runaway state of the power battery much faster. The threshold setting needs to be freely adjusted according to the situation and requirements. It can be determined by analyzing and verifying driving data from the power battery cloud big data platform to expand the scope of application and reduce the false alarm and missed alarm rates.

[0061] This invention also relates to a power battery thermal runaway early warning system based on singular spectral entropy. Corresponding to the aforementioned power battery thermal runaway early warning method based on singular spectral entropy, it can be understood as an implementation system for the power battery thermal runaway early warning method based on singular spectral entropy. It includes a module for constructing a pattern matrix, a singular value decomposition module, a module for calculating singular spectral entropy, and a normalization processing early warning analysis module connected in sequence. The pattern matrix construction module collects and uploads output signals from various sensors of the power battery from the vehicle end, constructing a pattern matrix according to the time-domain signal sequence formed by the time sequence of signal input. The singular value decomposition module performs singular value decomposition on the pattern matrix, arranging the singular values ​​to form the singular spectrum of the signal. The number of non-zero singular values ​​is equal to the total number of modes contained in each column of the mode matrix. The singular spectral entropy calculation module performs singular spectral analysis of the signal and calculates the proportion of each mode among all modes, thereby calculating the singular spectral entropy of the time-domain signal sequence. This reflects the uncertainty of each mode under the singular spectral division of the time-domain signal sequence and quantitatively describes the complex temporal morphological characteristics of the signal energy distribution. The normalization processing and early warning analysis module normalizes the obtained singular spectral entropy of the time-domain signal sequence based on a comparison with the singular spectral entropy of white noise. It then quantitatively evaluates the alarm coefficient based on the normalized singular spectral entropy and compares it with the alarm threshold to determine the risk of thermal runaway of the power battery and locate the fault location. In determining the thermal runaway of the power battery, the signals transmitted by the sensors are extracted and processed to find and analyze abnormal signals. This scheme analyzes the information content of the signal and provides early warning of thermal runaway by comparing the difference in information content between fault signals and normal signals. This greatly improves the system's sensitivity to fault signals, enabling faster detection of the moment of risk and location of the faulty unit. It effectively advances the time of thermal runaway warning and greatly improves safety.

[0062] Preferably, the pattern matrix construction module constructs the pattern matrix by setting the analysis window length and the time delay constant and extracting the time domain signal sequence in window order.

[0063] Preferably, the singular spectrum entropy calculation module calculates the information entropy of the singular spectrum based on the proportion of each mode in all modes, according to the definition of Shannon entropy, thereby obtaining the singular spectrum entropy of the time-domain signal sequence.

[0064] Preferably, the normalization processing early warning analysis module uses the Z-score method to standardize the normalized singular spectrum entropy. The difference between the normalized singular spectrum entropy of a certain power battery cell and the average of the normalized singular spectrum entropies of all power battery cells is divided by the standard deviation of the normalized singular spectrum entropies of all power battery cells to obtain the alarm coefficient of each power battery cell. The alarm coefficient of each power battery cell is compared with the alarm threshold to determine the thermal runaway risk of the power battery and locate the fault location.

[0065] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail with reference to the accompanying drawings and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention. In short, all technical solutions and improvements that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention patent.

Claims

1. A method for early warning of thermal runaway in power batteries based on singular spectral entropy, characterized in that, Includes the following steps: The process of constructing the pattern matrix involves collecting and uploading the output signals of various sensors in multiple battery cells of the power battery during vehicle operation or charging, and constructing the pattern matrix according to the time-domain signal sequence formed by the time sequence of signal input. The singular value decomposition step involves performing singular value decomposition on the mode matrix of each battery cell, arranging the singular values ​​to form the singular spectrum of the signal, and the number of non-zero singular values ​​is equal to the total number of modes contained in each column of the mode matrix. The steps for calculating singular spectral entropy involve performing singular spectral analysis on the signal and calculating the proportion of each mode in all modes, thereby calculating the singular spectral entropy of the time-domain signal sequence to reflect the uncertainty of each mode under the singular spectral division of the time-domain signal sequence, as well as to quantitatively describe the complex temporal morphological characteristics of the signal energy distribution. The normalization processing early warning analysis step involves normalizing the singular spectral entropy of the obtained time-domain signal sequence based on a comparison with the singular spectral entropy of white noise. The alarm coefficient is then quantitatively evaluated based on the normalized singular spectral entropy and compared with the alarm threshold. This includes standardizing the normalized singular spectral entropy using the Z-score method, dividing the difference between the normalized singular spectral entropy of a specific power battery cell and the average of the normalized singular spectral entropies of all power battery cells by the standard deviation of the normalized singular spectral entropies of all power battery cells to obtain the alarm coefficient for each power battery cell. The alarm coefficient of each power battery cell is then compared with the alarm threshold to determine the risk of thermal runaway of the power battery and locate the faulty power battery cell.

2. The method for early warning of thermal runaway in a power battery according to claim 1, characterized in that, In the step of constructing the pattern matrix, the pattern matrix is ​​constructed by setting the analysis window length and the time delay constant and extracting the time domain signal sequence in window order.

3. The method for early warning of thermal runaway of a power battery according to claim 1 or 2, characterized in that, In the step of calculating singular spectral entropy, based on the proportion of each mode in all modes, the information entropy of the singular spectrum is calculated according to the definition of Shannon entropy, thereby obtaining the singular spectral entropy of the time-domain signal sequence.

4. The method for early warning of thermal runaway of a power battery according to claim 1, characterized in that, In the step of constructing the mode matrix, the output signals of each sensor of the power battery include voltage, current, and temperature signals under several actual operating conditions or charging conditions when the vehicle is in motion.

5. The method for early warning of thermal runaway of a power battery according to claim 1, characterized in that, In the normalization processing early warning analysis step, the driving data from the power battery cloud big data platform is analyzed, verified, and adjusted to determine the alarm threshold range.

6. A power battery thermal runaway early warning system based on singular spectral entropy, characterized in that, It includes a module for constructing a pattern matrix, a module for singular value decomposition, a module for calculating singular spectral entropy, and a module for normalization processing and early warning analysis, all connected in sequence. The pattern matrix construction module collects and uploads the output signals of each sensor in multiple battery cells of the power battery under the vehicle driving or charging conditions, and constructs the pattern matrix according to the time domain signal sequence formed by the time sequence of signal input. The singular value decomposition module performs singular value decomposition on the mode matrix of each battery cell, arranges the singular values ​​to form the singular spectrum of the signal, and the number of non-zero singular values ​​is the total number of modes contained in each column of the mode matrix. The module for calculating singular spectral entropy performs singular spectral analysis of the signal and calculates the proportion of each mode in all modes, thereby calculating the singular spectral entropy of the time-domain signal sequence to reflect the uncertainty of each mode under the singular spectral division of the time-domain signal sequence, and to quantitatively describe the complex temporal morphological characteristics of the signal energy distribution. The normalization processing early warning analysis module normalizes the singular spectral entropy of the obtained time-domain signal sequence based on a comparison with the singular spectral entropy of white noise. It then quantitatively evaluates the alarm coefficient based on the normalized singular spectral entropy and compares it with the alarm threshold. This includes standardizing the normalized singular spectral entropy using the Z-score method, dividing the difference between the normalized singular spectral entropy of a specific power battery cell and the average of the normalized singular spectral entropies of all power battery cells by the standard deviation of the normalized singular spectral entropies of all power battery cells to obtain the alarm coefficient for each power battery cell. Finally, it compares the alarm coefficient of each power battery cell with the alarm threshold to determine the risk of thermal runaway of the power battery and locate the faulty power battery cell.

7. The power battery thermal runaway early warning system according to claim 6, characterized in that, The pattern matrix construction module constructs the pattern matrix by setting the analysis window length and time delay constant and extracting time-domain signal sequences in window order.

8. The power battery thermal runaway early warning system according to claim 6 or 7, characterized in that, The module for calculating singular spectral entropy calculates the information entropy of the singular spectrum based on the proportion of each mode in all modes and according to the definition of Shannon entropy, thereby obtaining the singular spectral entropy of the time-domain signal sequence.

Images