Multi-parameter coupling monitoring and early warning method for identifying thermal runaway of battery

Abstract

The invention discloses a multi-parameter coupling monitoring and early warning method for identifying thermal runaway of a battery, which belongs to the technical field of battery safety monitoring, and has the advantages that the early warning accuracy is obviously improved, and the false alarm rate and the missing report rate are obviously reduced compared with the traditional single-parameter method through multi-parameter coupling analysis; compared with a traditional method, the early warning time is remarkably increased in advance, meanwhile, the recognition accuracy under different working conditions (such as extreme temperature, overcharge and overdischarge) is effectively improved, the robustness and the anti-jamming capability are enhanced, multi-dimensional parameter cross validation is adopted, the influence of environmental noise is effectively reduced, and through the redundancy design of the optical fiber sensing network, the recognition accuracy is greatly improved. Even if part of the sensors fail, the system can still work stably, so that within a certain temperature range value, the fluctuation of early warning accuracy is small, and the system has good suppression capability on external interferences such as electromagnetic interference and mechanical vibration.

Description

Technical Field

[0001] The present invention relates to the technical field of battery safety monitoring, and in particular to a multi-parameter coupled monitoring and early warning method for identifying battery thermal runaway. Background Art

[0002] During the charging and discharging process, lithium-ion batteries may cause thermal runaway due to various factors, resulting in a sharp increase in temperature and triggering a chain reaction, ultimately causing serious safety accidents such as combustion and explosion. The existing monitoring methods mainly rely on single parameters such as temperature and voltage for judgment, and fail to fully consider the coupling relationship between parameters. The existing algorithms are not sensitive enough to the characteristics of the precursor stage of thermal runaway, which leads to delayed warning. There is a lack of systematic research on the evolution law of thermal runaway under different working conditions and different failure modes. As a result, the adopted warning algorithm lacks adaptability and generalization ability, and is difficult to adapt to complex working conditions. Therefore, a multi-parameter coupled monitoring and early warning method for identifying battery thermal runaway is proposed. Summary of the invention

[0003] The purpose of the present invention is to provide a multi-parameter coupled monitoring and early warning method for identifying battery thermal runaway, so as to solve the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solution: a multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway, comprising the following steps:

[0005] S1, multi-parameter data acquisition and preprocessing;

[0006] S2, multi-parameter coupling feature extraction;

[0007] S3. Establish a deep learning early warning model;

[0008] S4. Early warning strategy analysis.

[0009] As a further preferred embodiment of the present technical solution: in S1, the operation of multi-parameter data acquisition and preprocessing includes the following steps:

[0010] A1. Parameter acquisition system design;

[0011] A2. Data preprocessing;

[0012] In A1, the parameter acquisition system design includes temperature field monitoring, stress strain monitoring, electrochemical parameters and gas monitoring.

[0013] As a further preferred embodiment of the present technical solution: the temperature field monitoring adopts a fiber grating sensor array to measure the temperature field distribution on the battery surface, and the sampling frequency is not less than 10 Hz, and the stress strain monitoring arranges a strain gauge array to monitor the strain evolution of the battery shell.

[0014] As a further preferred embodiment of the present technical solution: the electrochemical parameters include voltage, current and AC impedance spectrum parameters with high precision, and the gas monitoring uses a VOC gas sensor, a CO gas sensor or a CO2 gas sensor.

[0015] As a further preferred embodiment of the present technical solution: in A2, data preprocessing is used to preprocess the collected raw data, which includes wavelet transform denoising, outlier detection and elimination, data standardization processing and time series alignment;

[0016] Among them, the mathematical model of data preprocessing is:

[0017] Wavelet transform denoising:

[0018] Assuming the original signal is f(t) and the wavelet basis function is ψj,k(t), the wavelet decomposition of the signal is:

[0019]

[0020] In formula (1), cj,k is the wavelet coefficient, and denoising is performed using a soft threshold function. The formula is expressed as:

[0021]

[0022] Data standardization:

[0023] For parameter x, the normalization process is:

[0024]

[0025] In formula (3), μ is the mean and σ is the standard deviation.

[0026] As a further preferred embodiment of the present technical solution: in S2, the operation of multi-parameter coupling feature extraction includes the following steps:

[0027] B1, time domain feature extraction;

[0028] B2, frequency domain feature extraction;

[0029] B3, multi-parameter coupling characteristics;

[0030] In B1, time domain feature extraction includes statistical features, trend features and morphological features.

[0031] As a further preferred embodiment of the present technical solution: the statistical characteristics include mean, variance, skewness and kurtosis, the trend characteristics include slope and acceleration, and the morphological characteristics include peaks, troughs and zero-crossing rate.

[0032] As a further preferred embodiment of the present technical solution: in B2, the frequency domain feature extraction is performed by Fourier transform to extract the frequency domain feature, and the formula thereof is expressed as follows:

[0033]

[0034] Calculate the power spectral density, the formula is expressed as:

[0035]

[0036] In B3, the coupling relationship matrix R between parameters is constructed, and its formula is expressed as:

[0037]

[0038] In formula (6), X i ,X j are time series with different parameters.

[0039] As a further preferred embodiment of the present technical solution: in S3, the operation of establishing the deep learning early warning model includes the following steps:

[0040] C1. Network structure design;

[0041] C2, model training and optimization;

[0042] In C1, the LSTM-Attention architecture is used, where the LSTM layer extracts temporal features, its attention mechanism weights key features, and the fully connected layer outputs warning results;

[0043] The forward propagation process of the LSTM unit is expressed as follows:

[0044] f t =σ(W f ·[h t-1 ,x t ]+b f )(7)

[0045] i t =σ(W i ·[h t-1 ,x t ]+b i )(8)

[0046]

[0047] o t =σ(W o [h t-1 ,x t ]+b o )(11)

[0048] h t =o t *tanh(C t )(12)

[0049] The attention weight calculation formula is expressed as:

[0050]

[0051] As a further preferred embodiment of the present technical solution: in C2, the cross entropy loss function Adam optimization algorithm is adopted, and its early stopping strategy avoids overfitting;

[0052] The loss function definition formula is expressed as:

[0053]

[0054] In formula (15), y ij is the true label, is the predicted value.

[0055] As a further preferred embodiment of the present technical solution: in S4, the operation of early warning strategy analysis includes the following steps:

[0056] D1. Risk level classification;

[0057] D2, early warning criteria;

[0058] In D1, the risk is thermal runaway risk, and the thermal runaway risk is divided into four levels:

[0059] Normal (Level 0), slightly abnormal (Level 1), moderately abnormal (Level 2), and severely abnormal (Level 3).

[0060] As a further preferred embodiment of the present technical solution: in D2, the warning criterion is based on the multi-parameter coupling characteristic value and the warning threshold, and its formula is expressed as follows:

[0061]

[0062] In formula (16), F is the coupling eigenvalue, θ i is the corresponding threshold.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. The accuracy of the early warning of the present invention is significantly improved. Through multi-parameter coupling analysis, the false alarm rate and missed alarm rate are significantly reduced compared with the traditional single-parameter method, and the early warning time is significantly increased in advance compared with the traditional method. At the same time, the recognition accuracy under different working conditions (such as extreme temperature, overcharging, over-discharging, etc.) is effectively improved;

[0065] 2. The robustness and anti-interference ability of the present invention are enhanced, and multi-dimensional parameter cross-validation is adopted to effectively reduce the influence of environmental noise. Moreover, through the redundant design of the optical fiber sensor network, even if some sensors fail, the system can still maintain stable operation, so that within a certain temperature range, the fluctuation of the early warning accuracy is small, and the present invention has good suppression ability against external interference such as electromagnetic interference and mechanical vibration;

[0066] 3. The present invention has strong adaptability and is suitable for different types of lithium-ion batteries (lithium iron phosphate, ternary materials, etc.). It has good mobility and good adaptability to different capacity levels (from single cells to battery packs). It also supports rapid deployment of different application scenarios (electric vehicles, energy storage power stations, etc.). The model can be continuously optimized through incremental learning, and the adaptability improves with the use time.

[0067] 4. The present invention has high computational efficiency, and the optimized algorithm structure greatly shortens the single warning judgment time, and has low resource occupancy. It can run in real time on embedded systems, and supports distributed deployment and parallel computing. It can monitor multiple battery systems at the same time, thereby effectively improving data processing efficiency;

[0068] 5. The present invention is highly practical, with clear classification of warning information, which is convenient for operation and maintenance personnel to respond quickly, supports remote monitoring and warning push, improves management efficiency, has historical data analysis and trend prediction functions, and is helpful for preventive maintenance. Based on the feature extraction method of multi-parameter coupling, the attention mechanism is innovatively applied to thermal runaway warning, which improves the model performance;

[0069] 6. The safety of the present invention is significantly improved, and early warning of battery thermal runaway can be achieved, which can gain sufficient time for emergency response. The set hierarchical warning mechanism can effectively prevent major safety accidents. The redundant design of the present invention ensures reliable operation in key scenarios, and has a self-diagnosis function, which can detect system abnormalities in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A schematic diagram of a flow chart of a multi-parameter coupled monitoring and early warning method for identifying battery thermal runaway according to the present invention;

[0071] Figure 2 It is a schematic diagram of the process of multi-parameter data collection and preprocessing in a multi-parameter coupled monitoring and early warning method for identifying battery thermal runaway of the present invention;

[0072] Figure 3 It is a schematic diagram of a process of extracting multi-parameter coupling features in a multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to the present invention;

[0073] Figure 4A schematic diagram of a process for establishing a deep learning early warning model in a multi-parameter coupling monitoring and early warning method for identifying thermal runaway of a battery according to the present invention;

[0074] Figure 5 The present invention is a schematic flow chart of early warning strategy analysis in a multi-parameter coupled monitoring and early warning method for identifying battery thermal runaway. DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0076] Example 1

[0077] See also Figure 1-Figure 5 The present invention provides a technical solution: a multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway, comprising the following steps:

[0078] S1, multi-parameter data acquisition and preprocessing;

[0079] S2, multi-parameter coupling feature extraction;

[0080] S3. Establish a deep learning early warning model;

[0081] S4. Early warning strategy analysis.

[0082] In this embodiment, specifically: in S1, the operation of multi-parameter data collection and preprocessing includes the following steps:

[0083] A1. Parameter acquisition system design;

[0084] A2. Data preprocessing;

[0085] In A1, the parameter acquisition system design includes temperature field monitoring, stress strain monitoring, electrochemical parameters and gas monitoring.

[0086] In this embodiment, specifically: the temperature field monitoring uses a fiber grating sensor array to measure the temperature field distribution on the battery surface, and the sampling frequency is not less than 10 Hz, and the stress and strain monitoring uses a strain gauge array to monitor the strain evolution of the battery shell.

[0087] In this embodiment, specifically: electrochemical parameters such as voltage, current and AC impedance spectrum are collected with high precision, and gas monitoring uses a variety of gas sensors, including VOC gas sensors, CO gas sensors and CO2 gas sensors.

[0088] In this embodiment, specifically: in A2, data preprocessing is used to preprocess the collected raw data, which includes wavelet transform denoising, outlier detection and elimination, data standardization processing and time series alignment;

[0089] Among them, the mathematical model of data preprocessing is:

[0090] Wavelet transform denoising:

[0091] Assuming the original signal is f(t) and the wavelet basis function is ψj,k(t), the wavelet decomposition of the signal is:

[0092]

[0093] In formula (1), cj,k is the wavelet coefficient, and denoising is performed using a soft threshold function. The formula is expressed as:

[0094]

[0095] Data standardization:

[0096] For parameter x, the normalization process is:

[0097]

[0098] In formula (3), μ is the mean and σ is the standard deviation.

[0099] In this embodiment, specifically: in S2, the operation of multi-parameter coupling feature extraction includes the following steps:

[0100] B1, time domain feature extraction;

[0101] B2, frequency domain feature extraction;

[0102] B3, multi-parameter coupling characteristics;

[0103] In B1, time domain feature extraction includes statistical features, trend features and morphological features.

[0104] In this embodiment, specifically: statistical characteristics include mean, variance, skewness and kurtosis, etc., trend characteristics include slope and acceleration, etc., and morphological characteristics include peaks, troughs and zero-crossing rate, etc.

[0105] In this embodiment, specifically: in B2, frequency domain feature extraction extracts frequency domain features by Fourier transform, and its formula is expressed as:

[0106]

[0107] Calculate the power spectral density, the formula is expressed as:

[0108]

[0109] In B3, the coupling relationship matrix R between parameters is constructed, and its formula is expressed as:

[0110]

[0111] In formula (6), X i ,X j are time series with different parameters.

[0112] In this embodiment, specifically: in S3, the operation of the deep learning early warning model is established, including the following steps:

[0113] C1. Network structure design;

[0114] C2, model training and optimization;

[0115] In C1, the LSTM-Attention architecture is used, where the LSTM layer extracts temporal features, its attention mechanism weights key features, and the fully connected layer outputs warning results;

[0116] The forward propagation process of the LSTM unit is expressed as follows:

[0117] f t =σ(W f ·[h t-1 ,x t ]+b f )(7)

[0118] i t =σ(W i ·[h t-1 ,x t ]+b i )(8)

[0119]

[0120] o t =σ(W o [h t-1 ,x t ]+b o )(11)

[0121] h t =o t *tanh(C t )(12)

[0122] The attention weight calculation formula is expressed as:

[0123]

[0124] In this embodiment, specifically: in C2, the cross entropy loss function Adam optimization algorithm is adopted, and its early stopping strategy avoids overfitting;

[0125] The loss function definition formula is expressed as:

[0126]

[0127] In formula (15), y ij is the true label, is the predicted value.

[0128] In this embodiment, specifically: in S4, the operation of early warning strategy analysis includes the following steps:

[0129] D1. Risk level classification;

[0130] D2, early warning criteria;

[0131] In D1, the risk is thermal runaway risk, and the thermal runaway risk is divided into four levels:

[0132] Normal (Level 0), slightly abnormal (Level 1), moderately abnormal (Level 2), and severely abnormal (Level 3).

[0133] In this embodiment, specifically: in D2, the warning criterion is based on the multi-parameter coupling characteristic value and the warning threshold, and its formula is expressed as follows:

[0134]

[0135] In formula (16), F is the coupling eigenvalue, θ i is the corresponding threshold.

[0136] Example 2

[0137] For thermal runaway warning of single battery:

[0138] P1. System configuration;

[0139] P11. Hardware Configuration

[0140] Pressure and temperature measurement: using "temperature + pressure" integrated flexible pressure film;

[0141] Gas measurement: MEMS gas sensor array sensors such as CO, H2, CO2, VOC, pyrolysis particles, etc.

[0142] Voltage and current measurement: collect BMS battery management system data;

[0143] P12. Sampling parameters

[0144] Temperature sampling: 10Hz, strain sampling: 100Hz, electrical parameters: 1kHz, gas concentration: 1Hz;

[0145] P2, feature engineering;

[0146] P21. Time domain feature calculation

[0147] Calculate feature parameters within a sliding time window:

[0148] Temperature gradient:

[0149]

[0150] Strain Energy:

[0151]

[0152] Voltage fluctuations:

[0153]

[0154] P22. Frequency domain feature extraction

[0155] Including power spectrum density calculation, main frequency component extraction and band capacity ratio calculation;

[0156] P3, early warning model implementation;

[0157] P31. Network Configuration

[0158] LSTM layer: 128 hidden units and attention layer, 8-head self-attention mechanism, fully connected layer: using 256-128-64-4 structure;

[0159] P32. Training parameters

[0160] Batch size: 64, learning rate: 0.001, training epochs: 100, validation ratio: 20%.

[0161] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway, characterized in that: The following steps are involved: S1, multi-parameter data acquisition and preprocessing; S2, multi-parameter coupling feature extraction; S3. Establish a deep learning early warning model; S4. Early warning strategy analysis.

2. A multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to claim 1, characterized in that: In S1, the multi-parameter data acquisition and preprocessing operation includes the following steps: A1. Parameter acquisition system design; A2. Data preprocessing; In A1, the parameter acquisition system design includes temperature field monitoring, stress strain monitoring, electrochemical parameters and gas monitoring.

3. A multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to claim 2, characterized in that: In A2, data preprocessing is used to preprocess the collected raw data, which includes wavelet transform denoising, outlier detection and elimination, data standardization and time series alignment; Among them, the mathematical model of data preprocessing is: Wavelet transform denoising: Assuming the original signal is f(t) and the wavelet basis function is ψj,k(t), the wavelet decomposition of the signal is: In formula (1), cj,k is the wavelet coefficient, and denoising is performed using a soft threshold function. The formula is expressed as: Data standardization: For parameter x, the normalization process is: In formula (3), μ is the mean and σ is the standard deviation.

4. The multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to claim 1, characterized in that: In S2, the operation of multi-parameter coupling feature extraction includes the following steps: B1, time domain feature extraction; B2, frequency domain feature extraction; B3, multi-parameter coupling characteristics; In B1, time domain feature extraction includes statistical features, trend features and morphological features.

5. A multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to claim 4, characterized in that: In B2, frequency domain feature extraction extracts frequency domain features through Fourier transform, and its formula is expressed as: Calculate the power spectral density, the formula is expressed as: In B3, the coupling relationship matrix R between parameters is constructed, and its formula is expressed as: In formula (6), X i ,X j are time series with different parameters.

6. The multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to claim 1, characterized in that: In S3, the operation of the deep learning early warning model is established, including the following steps: C1. Network structure design; C2, model training and optimization; In C1, the LSTM-Attention architecture is used, where the LSTM layer extracts temporal features, its attention mechanism weights key features, and the fully connected layer outputs warning results; The forward propagation process of the LSTM unit is expressed as follows: f t =σ(W f ·[h t-1 ,x t ]+b f )(7) i t =σ(W i ·[h t-1 ,x t ]+b i )(8) the t =σ(W o [h t-1 ,x t ]+b o (11) h t =o t *tanh(C t )(12) The attention weight calculation formula is expressed as:

7. A multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to claim 6, characterized in that: In C2, the cross entropy loss function Adam optimization algorithm is used, and its early stopping strategy avoids overfitting; The loss function definition formula is expressed as: In formula (15), y ij is the true label, is the predicted value.

8. The multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to claim 1, characterized in that: In S4, the operation of early warning strategy analysis includes the following steps: D1. Risk level classification; D2, early warning criteria; In D1, the risk is thermal runaway risk, and the thermal runaway risk is divided into four levels: Normal (Level 0), slightly abnormal (Level 1), moderately abnormal (Level 2), and severely abnormal (Level 3).

9. A multi-parameter coupling monitoring and early warning method for identifying battery thermal runaway according to claim 8, characterized in that: In D2, the early warning criterion is based on the multi-parameter coupling characteristic value and the early warning threshold, and its formula is expressed as follows: In formula (16), F is the coupling eigenvalue, θ i is the corresponding threshold.

Images