Method and System for Thermal Runaway Early Warning of Lithium-Ion Batteries Based on Multimodal Reconstruction and Fusion

Abstract

The present disclosure provides a method and system for early warning of thermal runaway of lithium-ion batteries based on multimodal reconstruction and fusion, which relates to the technical field of early warning of thermal runaway of lithium-ion batteries. The method includes obtaining the time-domain signal of the sound of the lithium-ion battery, extracting the time-domain characteristics and frequency-domain characteristics of the sound of battery thermal runaway by using time-frequency transformation and phase space reconstruction, then processing the time-domain and frequency-domain characteristics through GRU and squeeze-and-excitation attention mechanism respectively, and finally realizing the recognition of the sound signal through cross-modal attention fusion and linear classifier. The present disclosure can effectively improve the safety of lithium-ion batteries, timely warn of the risk of thermal runaway, and avoid potential safety hazards.

Description

Technical Field

[0001] The present disclosure relates to the technical field of early warning of thermal runaway of lithium-ion batteries, and specifically relates to a method and system for early warning of thermal runaway of lithium-ion batteries based on multimodal reconstruction fusion. Background Technique

[0002] The statements in this part only provide background technical information related to the present disclosure and do not necessarily constitute prior art.

[0003] Lithium-ion batteries, with their high energy density, long cycle life, and stable charge and discharge performance, have become an indispensable energy storage device in fields such as smartphones, electric vehicles, and renewable energy storage, and are of great significance for improving equipment efficiency and promoting the transformation of green energy. However, under extreme conditions such as overcharging, high temperature, or physical damage, lithium-ion batteries have a risk of thermal runaway. This phenomenon is accompanied by a sharp rise in the internal temperature of the battery, which may trigger the decomposition of the electrolyte, gas expansion, and even short circuits, ultimately leading to serious safety threats such as fires or explosions, posing major potential hazards to personnel safety and equipment operation.

[0004] Currently, the methods for early warning of thermal runaway of lithium-ion batteries mainly rely on signals such as temperature, voltage, current, and gas detection. However, these methods have problems such as response lag, high cost, or complex deployment. For example, in the invention patent - CN202411256611.2, by coupling the state of charge, health state, thermodynamics, and chemical reaction models of the battery, the total heat generation rate and temperature change are calculated to predict the thermal runaway situation. However, this method still needs to be optimized for the complexity of the model and its adaptability to different working conditions. The invention patent - CN202410947610.6 proposes a method and system for early warning of thermal runaway based on battery gas detection. However, this method has high deployment requirements and the cost of gas sensors is high. In contrast, acoustic signals have the advantages of low cost and non-contact monitoring, and show great potential in the early warning of thermal runaway of lithium-ion batteries.

[0005] However, acoustic signals are easily interfered by noise in a complex environment and cannot effectively extract key features. Before obvious changes occur in traditional monitoring indicators such as voltage, current, and temperature, it is impossible to achieve high-sensitivity early warning of the thermal runaway situation of lithium-ion batteries. In addition, there are some high-frequency fluctuations or anomalies in acoustic signals, and existing deep learning models cannot comprehensively capture the complex fluctuating features during the thermal runaway process of the battery, resulting in low prediction accuracy. Summary of the Invention

[0006] To solve the above problems, the present disclosure proposes a method and system for early warning of lithium-ion battery thermal runaway based on multi-modal reconstruction and fusion. The method extracts the characteristics of the battery thermal runaway sound signal through time-frequency transformation and phase space reconstruction, processes the time-domain and frequency-domain characteristics through a Gated Recurrent Unit (GRU) and a Squeeze-and-Excitation (SE) respectively, and finally realizes the identification of fault signals through cross-modal attention fusion and a linear classifier, effectively improving the safety of lithium-ion batteries, timely warning of thermal runaway risks, and avoiding potential safety hazards.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] A method for early warning of lithium-ion battery thermal runaway based on multi-modal reconstruction and fusion, comprising:

[0009] Obtain the time-domain sound signal of the lithium-ion battery, and apply Fourier transform to convert the time-domain sound signal into a frequency-domain sound signal;

[0010] Based on the phase space reconstruction method, determine the time delay and embedding dimension, perform high-dimensional embedding on the time-domain sound signal and the frequency-domain sound signal respectively to reconstruct the high-dimensional phase space, and extract the phase space reconstruction characteristics of the time-domain sound signal and the frequency-domain sound signal in the high-dimensional phase space respectively, to obtain time-domain characteristics and frequency-domain characteristics;

[0011] Input the time-domain characteristics into the gated recurrent unit for time modeling, and input the frequency-domain characteristics into the squeeze-and-excitation attention module for weighted processing to obtain weighted frequency-domain characteristics;

[0012] Calculate the relationship between the weighted frequency-domain characteristics and the time-modeled time-domain characteristics through a cross-modal attention mechanism, dynamically fuse the weighted frequency-domain characteristics and the time-modeled time-domain characteristics to obtain fusion characteristics, input the fusion characteristics into the fully connected layer, and through linear transformation and activation function, output the classification result of lithium-ion battery thermal runaway.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] A system for early warning of lithium-ion battery thermal runaway based on multi-modal reconstruction and fusion, comprising:

[0015] A signal acquisition module, configured to acquire the time-domain sound signal of the lithium-ion battery, and apply Fourier transform to convert the time-domain sound signal into a frequency-domain sound signal;

[0016] A feature extraction module, which is used to determine the time delay and embedding dimension based on the phase space reconstruction method, perform high-dimensional embedding on the sound time-domain signal and the sound frequency-domain signal respectively to reconstruct the high-dimensional phase space, and extract the phase space reconstruction features of the sound time-domain signal and the sound frequency-domain signal in the high-dimensional phase space respectively, so as to obtain the time-domain features and the frequency-domain features;

[0017] A feature fusion module, which is used to input the time-domain features into a gated recurrent unit for time modeling, input the frequency-domain features into a squeeze-and-excitation attention module for weighted processing to obtain weighted frequency-domain features; calculate the relationship between the weighted frequency-domain features and the time-modeled time-domain features through a cross-modal attention mechanism, and dynamically fuse the weighted frequency-domain features and the time-modeled time-domain features to obtain fusion features;

[0018] A classification module, which is used to input the fusion features into a fully connected layer, and through linear transformation and activation function, output the classification result of the thermal runaway of the lithium-ion battery.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A computer program product, including a computer program, which when executed by a processor, implements the method for early warning of thermal runaway of a lithium-ion battery based on multi-modal reconstruction and fusion.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] A non-transitory computer-readable storage medium, which is used to store computer instructions, and when the computer instructions are executed by a processor, the method for early warning of thermal runaway of a lithium-ion battery based on multi-modal reconstruction and fusion is implemented.

[0023] According to some embodiments, the present disclosure adopts the following technical solutions:

[0024] An electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory, so that the electronic device executes and implements the method for early warning of thermal runaway of a lithium-ion battery based on multi-modal reconstruction and fusion.

[0025] Compared with the prior art, the beneficial effects of the present disclosure are:

[0026] The thermal runaway early warning method for lithium-ion batteries based on multi-modal reconstruction and fusion disclosed in the present disclosure. During the thermal runaway of lithium-ion batteries, there are mainly three main sound signal types: noise, thermal runaway expansion sound, and pseudo-expansion sound. By using the unique sound characteristics of acoustic information for diagnosis, it can early and highly sensitively warn of the thermal runaway situation of lithium-ion batteries before obvious changes appear in traditional monitoring indicators such as voltage, current, and temperature.

[0027] The thermal runaway early warning method for lithium-ion batteries based on multi-modal reconstruction and fusion disclosed in the present disclosure. Since noise usually has random and wide spectral characteristics with large frequency variations, time-frequency transformation technology is used to decompose and extract noise and target signals from time-frequency signals. Through phase space reconstruction, high-dimensional embedding of time-domain and frequency-domain signals is performed to capture the short-term dynamics and frequency characteristics of the signals, thereby accurately extracting the time change patterns and spectral fluctuation characteristics in the battery sound signals, especially the unique characteristics of thermal runaway expansion sound and noise, providing a comprehensive description for effectively identifying the thermal runaway phenomenon.

[0028] The thermal runaway early warning method for lithium-ion batteries based on multi-modal reconstruction and fusion disclosed in the present disclosure. A thermal runaway sound classifier model with an extrusion excitation attention mechanism and a cross-modal attention mechanism is established according to the sound characteristics of lithium-ion batteries to classify and identify noises such as noise, thermal runaway expansion sound, and pseudo-expansion sound, ensuring that the thermal runaway event can trigger an early warning in a timely manner and having a good thermal runaway recognition rate.

[0029] The thermal runaway early warning method and system for lithium-ion batteries based on multi-modal reconstruction and fusion disclosed in the present disclosure. Compared with traditional methods, its implementation does not depend on high-cost hardware device configurations, thus effectively reducing the investment cost of the early warning system. In addition, the present disclosure takes sound signal analysis as the core, has high flexibility and adaptability, and can successfully deploy and implement the early warning function in different application scenarios, providing a new solution for the safety management of lithium-ion batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The specification drawings forming a part of the present disclosure are used to provide a further understanding of the present disclosure. The schematic embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation of the present disclosure.

[0031] Figure 1 It is a flowchart of the implementation of the thermal runaway early warning method for lithium-ion batteries based on multi-modal reconstruction and fusion according to an embodiment of the present disclosure;

[0032] Figure 2 It is a sound waveform diagram of the 125% SOC thermal runaway experiment according to an embodiment of the present disclosure;

[0033] Figure 3 It is a confusion matrix of the sound classifier according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] The present disclosure will be further described below in conjunction with the accompanying drawings and embodiments.

[0035] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present disclosure belongs.

[0036] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0037] Embodiment 1

[0038] In an embodiment of the present disclosure, a method for predicting thermal runaway of a lithium-ion battery based on multimodal reconstruction and fusion is provided. The method steps are as follows:

[0039] Step 1: Obtain the sound time-domain signal of the lithium-ion battery, and apply Fourier transform to convert the sound time-domain signal to obtain the sound frequency-domain signal;

[0040] Step 2: Based on the phase space reconstruction method, determine the time delay and embedding dimension, and perform high-dimensional embedding on the sound time-domain signal and the sound frequency-domain signal respectively to reconstruct the high-dimensional phase space. Extract the phase space reconstruction features of the sound time-domain signal and the sound frequency-domain signal in the high-dimensional phase space respectively to obtain the time-domain features and the frequency-domain features;

[0041] Step 3: Input the time-domain features into a gated recurrent unit for time modeling, and input the frequency-domain features into a squeeze-and-excitation attention module for weighting processing to obtain weighted frequency-domain features;

[0042] Step 4: Calculate the relationship between the weighted frequency-domain features and the time-modeled time-domain features through a cross-modal attention mechanism, dynamically fuse the weighted frequency-domain features and the time-modeled time-domain features to obtain fusion features, input the fusion features into a fully connected layer, and output the classification result of the thermal runaway of the lithium-ion battery through linear transformation and activation function.

[0043] As an embodiment, in the thermal runaway experiment of lithium-ion batteries, the thermal runaway warning method based on multimodal reconstruction fusion of the present disclosure mainly discovers three sound signals: noise, thermal runaway expansion sound, and noise similar to expansion sound. Noise usually has random and wide spectral characteristics with large frequency variations. Therefore, the present disclosure adopts time-frequency transformation technology, which can decompose time-frequency signals and extract noise and target signals. The thermal runaway expansion sound has a unique frequency pattern and exhibits obvious characteristics in both the time domain and the frequency domain. To effectively capture these characteristics, phase space reconstruction technology is used to transform time-domain and frequency-domain signals into high-dimensional signals. At the same time, a Gated Recurrent Unit (GRU) module is used for time-domain feature extraction and can process such progressive signals. The noise similar to expansion sound has similar amplitude characteristics to the expansion sound and is prone to causing interference. Therefore, a Squeeze-and-Excitation (SE) module is designed to weight the frequency-domain features, enhance the frequency components of the thermal runaway expansion sound, and suppress interference. Finally, a cross-modal attention mechanism fuses time-domain and frequency-domain features, and a linear classifier is used to classify and identify these three sound signals, effectively distinguishing noise, thermal runaway expansion sound, and noise similar to expansion sound, ensuring that the thermal runaway event can trigger a warning in a timely manner. The specific implementation process is as follows:

[0044] Step 1: Obtain the sound time-domain signal of the lithium-ion battery, and apply Fourier transform to convert the sound time-domain signal into a sound frequency-domain signal;

[0045] Specifically, install a sound sensor at a set distance from the center of the lithium-ion battery to collect the sound time-domain signal of the lithium-ion battery in real time.

[0046] Furthermore, apply Fourier transform to convert the sound time-domain signal into a sound frequency-domain signal. Specifically, Fourier transform converts the time-domain signal into a frequency-domain signal The conversion process is as follows:

[0047]

[0048] where, refers to the frequency-domain signal, refers to the time-domain signal, is the frequency variable, is the time, is the imaginary unit. Fourier transform obtains the corresponding frequency components by integrating and summing the time-domain signal over the entire time interval. The converted frequency-domain signal contains the amplitude and phase information of each frequency component and can reveal the distribution characteristics of the signal at different frequencies.

[0049] Step 2: Based on the phase space reconstruction method, determine the time delay and embedding dimension, and perform high-dimensional embedding on the sound time-domain signal and the sound frequency-domain signal respectively to reconstruct the high-dimensional phase space, and extract the phase space reconstruction features of the sound time-domain signal and the sound frequency-domain signal in the high-dimensional phase space respectively to obtain the time-domain features and frequency-domain features;

[0050] Specifically, based on the phase space reconstruction method, embed the sound time-domain signal and the sound frequency-domain signal into the high-dimensional phase space respectively to reconstruct the high-dimensional phase space. Among them, according to the time delay embedding theorem, determine the time delay and embedding dimension so that the reconstructed phase space retains the structure and its evolution characteristics of the original dynamic system.

[0051] First, determine the appropriate time delay. The calculation formula for the time delay is:

[0052] (1)

[0053] where, refers to the sound time-domain signal, refers to the current moment, refers to the sum of the current moment and the delay time, is the autocorrelation value at the time delay, is the average value of the time series, N is the total length of the time series. The first zero or the first local minimum of the autocorrelation function is usually used as the time delay.

[0054] Secondly, determine the embedding dimension, and determine the embedding dimension in the phase space reconstruction based on the false nearest neighbor method. By calculating the distances between adjacent points at different dimensions, judge whether there are false nearest neighbor points. The calculation process is:

[0055] (2)

[0056] where, m is the embedding dimension, is the time delay.

[0057] If the distance between adjacent points increases significantly after increasing the dimension, then this point is considered a false nearest neighbor. By calculating the proportion of false nearest neighbor points, select the appropriate embedding dimension m so that the error is minimized, and then effectively reconstruct the phase space of the original dynamic system.

[0058] The finally reconstructed trajectory matrix can be written as:

[0059] (3)

[0060] Furthermore, based on the reconstructed high-dimensional phase space, non-linear feature extraction is performed on the sound time-domain signal and the sound frequency-domain signal from different angles in a dual-channel manner. First, the phase space of the sound signal is reconstructed to obtain its inherent dynamic structure. Then, in the high-dimensional phase space of the time-domain signal and the frequency-domain signal, five non-linear features are extracted respectively, including approximate entropy, Shannon entropy, fractal coefficient, correlation dimension, and recurrence rate. Among them, approximate entropy and Shannon entropy are extracted using a calculation method based on probability distribution, the fractal coefficient is extracted based on the box-counting method or the Higuchi fractal dimension algorithm, the correlation dimension is extracted using the Grassberger-Procaccia algorithm, and the recurrence rate feature is extracted through recurrence plot analysis. The same five phase space reconstruction features from different angles are extracted for each signal. That is, the phase space reconstruction features of the sound time-domain signal are approximate entropy, Shannon entropy, fractal coefficient, correlation dimension, and recurrence rate feature, which are uniformly used as time-domain features; the phase space reconstruction features of the sound frequency-domain signal in the high-dimensional phase space are also approximate entropy, Shannon entropy, fractal coefficient, correlation dimension, and recurrence rate feature, which are uniformly used as frequency-domain features. As shown in Table 1, Table 1 shows the features extracted after phase space reconstruction.

[0061] Table 1 Features Extracted after Phase Space Reconstruction

[0062]

[0063] Step 3: Input the time-domain features into a gated recurrent unit for time modeling, and input the frequency-domain features into a squeeze-and-excitation attention module for weighted processing to obtain weighted frequency-domain features;

[0064] Specifically, after completing the extraction of phase space reconstruction features, the present disclosure further designs a deep learning framework that combines time-domain and frequency-domain features. The time-domain features and the frequency-domain features respectively reflect the dynamic response of the battery in the time dimension and the change of frequency components, and the two complement each other, jointly improving the prediction performance of the model in thermal runaway detection.

[0065] The time-domain features and the frequency-domain features are further input into a sound classifier model for processing, which includes the processing of a gated recurrent unit and channel attention (squeeze-and-excitation). First, the time-domain features are input into the gated recurrent unit, which is composed of a convolutional neural network and a GRU. The time-domain features are processed through the convolutional neural network (Convolutional Neural Network, CNN) and the GRU. The CNN can capture the local time dependence in the signal, while the GRU processes the long-term dependence of the thermal runaway sound signal of the lithium-ion battery through its dynamic time modeling ability.

[0066] Specifically, the time-domain features of the thermal runaway sound signals of lithium-ion batteries are processed by combining a convolutional neural network (CNN) with a gated recurrent unit (GRU). Among them, CNN is mainly used to extract local time-dependent features, while GRU captures long-term dependencies through its dynamic time modeling ability. First, two one-dimensional convolutional layers (Conv1D) are adopted. The use of a small receptive field (kernel size of 3) ensures the ability to capture feature patterns within short time series. At the same time, the ReLU activation function is used to enhance the non-linear expression ability and prevent the problem of gradient vanishing. In addition, the pooling layer further reduces the time series dimension, improves the computational efficiency, and extracts the globally most significant features. Subsequently, a bidirectional GRU (Bidirectional GRU) is used to simultaneously consider the dynamic change features in the forward and backward directions of the signal, thereby enhancing the modeling ability for complex time-dependent relationships. Finally, the time series features output by GRU are globally averaged to form a fixed-dimensional representation, ensuring that the model can capture both local short-time features and model the long-term evolution trend of the signal during learning, thereby effectively improving the classification accuracy of the thermal runaway sound signals of lithium-ion batteries.

[0067] On the other hand, the frequency-domain features are weighted by the SE module. The SE module first performs global average pooling on each feature channel to extract global information and generate a vector describing the channel importance. This vector then undergoes two fully connected networks for compression and reconstruction. The final output is used as the weight coefficient of the channel to weight the frequency-domain features, enabling the model to automatically focus on the key frequency components, as follows:

[0068] (4)

[0069] Among them, x is the input frequency-domain feature, W 1 and W 2 are the weight matrices in the network, 𝜎 is the Sigmoid activation function, and the output y is the weighted frequency-domain feature.

[0070] Step 4: Calculate the relationship between the weighted frequency-domain features and the time-domain features after time modeling through a cross-modal attention mechanism, dynamically fuse the weighted frequency-domain features and the time-domain features after time modeling to obtain fused features, and input the fused features into the fully connected layer. Through linear transformation and activation function, the classification result of the thermal runaway of lithium-ion batteries is output.

[0071] Specifically, calculate the relationship between the weighted frequency-domain features in the time domain and the time-domain features after time modeling through a cross-modal attention mechanism, and dynamically weight the features of each modality to achieve the organic fusion of information and obtain fused features. The mathematical formula of the cross-modal attention mechanism is:

[0072] Attention = softmax( QK T ) (5)

[0073] Fused_features = Attention× V (6)

[0074] Among them, the time-domain feature is used as Q and K , and the frequency-domain feature is used as V , Fused_features is the fusion feature, and T refers to the transpose of the K matrix.

[0075] The above method of the present disclosure is based on the complementarity of time-domain and frequency-domain features. The time-domain feature effectively captures the dynamic changes of the signal, reflects the mutations and instantaneous fluctuations in the thermal runaway process, and is therefore used to guide attention to the key changes in time. The frequency-domain feature reveals the frequency components of the signal, can identify high-frequency fluctuations and anomalies, provides stable frequency information as its value, and enhances the weighted effect of the time-domain feature. Through this cross-modal fusion, the model can more comprehensively capture the complex features in the battery thermal runaway process and improve the prediction accuracy.

[0076] Furthermore, the fusion feature is input into the fully connected layer. This layer finally outputs the classification result through further linear transformation and activation function. This process is achieved through the following formula:

[0077] Output =softmax fc ( Fused_features )] (7)

[0078] Among them, fc is the fully connected layer. The probability of each classification is output through the softmax function, and the final result of the classification is determined according to the maximum probability, thereby realizing the identification and early warning of lithium-ion battery thermal runaway. Thus, the pseudo-code of the thermal runaway sound classifier can be obtained as shown in Table 2.

[0079] Table 2 Thermal runaway sound classification algorithm framework

[0080]

[0081] Simulation experiment

[0082] The experimental platform of the present disclosure aims to study the thermal runaway characteristics of lithium-ion batteries. The core components include a soft-pack battery, a charge-discharge test instrument (Nebula 5V30A cell energy feedback system), a host computer, and an accelerating calorimeter. To comprehensively capture the information during the thermal runaway process, a sound sensor was arranged 20 centimeters away from the center of the battery in the experiment to monitor the acoustic signal in real time. Meanwhile, a camera was installed on the top cover of the accelerating calorimeter to ensure data synchronization and accurate recording of experimental phenomena. The present disclosure uses the C rate to describe the charging / discharging current of the battery, where the charging rate of dataset 1 is 3.5C (14A). SOC (State of Charge) represents the ratio of the current charge of the battery to the nominal capacity, reflects the remaining charge of the battery, and plays a key role in evaluating the battery performance and life. Dataset 2 was collected when the SOC was 125%, indicating that the battery was in an overcharged state. Dataset 3 was collected when the SOC was 110%.

[0083] Taking the experiment at 125% SOC as an example, Figure 2 shows the correlation between the sound signal and the battery state. By comparing the changes in the battery state at high-amplitude moments. Although the sound amplitude reached a peak at times t 1 、t 2 、t 3 and t 4 , the battery only swelled at time t 3 , while there were no corresponding changes at times t 1 、t 2 and t 4 . This indicates that high amplitude is not always directly related to battery swelling, and there may be noise interference similar to thermal runaway. For these three sound signals, the sound classifier based on phase space reconstruction and hybrid attention mechanism of the present disclosure was used for prediction and classification.

[0084] To verify the sound classifier of the present disclosure, 80% of the data obtained from the experiment was used as the training set, and the remaining 20% of the data was used as the test set. To eliminate accidental factors, the average value of 10 experiments was adopted. After the model training was completed, the confusion matrix results were as shown in Figure 3 . Figure 3 In it, 1 represents noise, 2 represents noise similar to the swelling sound, and 3 represents the swelling sound. The experimental results show that the sound classification algorithm proposed by the present disclosure can effectively handle the noise in battery sound classification, and the average recognition accuracy reaches 98.6%.

[0085] By mapping the features extracted by phase space reconstruction back to the original thermal runaway sound sequence collected, the thermal runaway occurrence time and the early warning time based on the sound signal can be obtained, as shown in Table 3. By analyzing the results of three experiments, it was found that the thermal runaway early warning time based on the sound signal could trigger the early warning about half of the thermal runaway occurrence time in advance.

[0086] Table 3 Thermal runaway warning time and occurrence time

[0087]

[0088] In summary, the thermal runaway warning method for lithium-ion batteries based on multimodal reconstruction fusion proposed in this disclosure integrates phase space reconstruction and hybrid attention mechanism. This warning method does not require a high-precision battery mathematical model, nor does it require a complex feature acquisition hardware system, effectively reducing the modeling cycle and hardware cost. This technology shows significant application potential and practical value in enhancing the safety of battery systems.

[0089] Embodiment 2

[0090] In an embodiment of the present disclosure, a thermal runaway warning system for lithium-ion batteries based on multimodal reconstruction fusion is provided, including:

[0091] A signal acquisition module, configured to acquire the sound time-domain signal of the lithium-ion battery, and apply Fourier transform to convert the sound time-domain signal to obtain the sound frequency-domain signal;

[0092] A feature extraction module, configured to determine the time delay and embedding dimension based on the phase space reconstruction method, perform high-dimensional embedding on the sound time-domain signal and the sound frequency-domain signal respectively to reconstruct the high-dimensional phase space, and extract the phase space reconstruction features of the sound time-domain signal and the sound frequency-domain signal in the high-dimensional phase space respectively to obtain the time-domain features and the frequency-domain features;

[0093] A feature fusion module, configured to input the time-domain features into a gated recurrent unit for time modeling, input the frequency-domain features into a squeeze-and-excitation attention module for weighting processing to obtain weighted frequency-domain features; calculate the relationship between the weighted frequency-domain features and the time-modeled time-domain features through a cross-modal attention mechanism, and dynamically fuse the weighted frequency-domain features and the time-modeled time-domain features to obtain fusion features;

[0094] A classification module, configured to input the fusion features into a fully connected layer, and output the thermal runaway classification result of the lithium-ion battery through linear transformation and activation function.

[0095] Embodiment 3

[0096] In an embodiment of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, it implements the thermal runaway warning method for lithium-ion batteries based on multimodal reconstruction fusion.

[0097] Embodiment 4

[0098] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for predicting thermal runaway of a lithium-ion battery based on multimodal reconstruction and fusion as described above is implemented.

[0099] Embodiment 5

[0100] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes the method for predicting thermal runaway of a lithium-ion battery based on multimodal reconstruction and fusion as described above.

[0101] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0103] Although the specific implementation manners of the present disclosure are described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that based on the technical solutions of the present disclosure, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present disclosure.

Claims

1. A lithium-ion battery thermal runaway early warning method based on multi-modal reconstruction fusion, characterized in that: include: Acquire the sound time domain signal of the lithium-ion battery, and transform the sound time domain signal into the sound frequency domain signal by Fourier transform; Based on the phase space reconstruction method, the time delay and embedding dimension are determined, and the sound time domain signal and the sound frequency domain signal are respectively embedded in high dimensions to reconstruct the high-dimensional phase space, and the phase space reconstruction features of the sound time domain signal and the sound frequency domain signal in the high-dimensional phase space are respectively extracted to obtain the time domain features and the frequency domain features; The time domain features are input into the gated recurrent unit for time modeling, and the frequency domain features are input into the squeeze-excitation attention module for weighted processing to obtain weighted frequency domain features; The relationship between the weighted frequency domain features and the time domain features after time modeling is calculated through the cross-modal attention mechanism, and the weighted frequency domain features and the time domain features after time modeling are dynamically fused to obtain the fused features. The fused features are input into the fully connected layer, and the thermal runaway classification results of lithium-ion batteries are output through linear transformation and activation function.

2. The lithium-ion battery thermal runaway early warning method based on multi-modal reconstruction fusion according to claim 1, characterized in that: Based on the phase space reconstruction method, the sound time domain signal and the sound frequency domain signal are embedded into the high-dimensional phase space respectively. According to the time delay embedding theorem, the time delay and embedding dimension are determined so that the reconstructed phase space retains the structure and evolution characteristics of the original dynamic system. The calculation process of time delay is: in, It refers to the sound time domain signal. Refers to the current moment, Refers to the sum of the current time and the delay time. is the autocorrelation value at time delay, is the mean value of the time series, N is the total length of the time series; the first zero or the first local minimum of the autocorrelation function is used as the time lag.

3. The lithium-ion battery thermal runaway early warning method based on multi-modal reconstruction fusion according to claim 1, characterized in that: The embedding dimension in phase space reconstruction is determined based on the pseudo-nearest neighbor method. By calculating the distance between adjacent points in different dimensions, it is determined whether there are pseudo-nearest neighbor points. The calculation process is as follows: in, m is the embedding dimension, It's a time delay.

4. The lithium-ion battery thermal runaway early warning method based on multi-modal reconstruction fusion according to claim 1, characterized in that: The phase space reconstruction features of the sound time domain signal and the acoustic frequency domain signal in the high-dimensional phase space are extracted respectively, including: extracting the phase space reconstruction features of the sound time domain signal in the high-dimensional phase space, the phase space reconstruction features of the sound time domain signal are approximate entropy, Shannon entropy, fractal coefficient, correlation dimension and recurrence rate features, which are unified as time domain features; extracting the phase space reconstruction features of the acoustic frequency domain signal in the high-dimensional phase space, the phase space reconstruction features of the acoustic frequency domain signal are also approximate entropy, Shannon entropy, fractal coefficient, correlation dimension and recurrence rate features, which are unified as frequency domain features.

5. The lithium-ion battery thermal runaway early warning method based on multi-modal reconstruction fusion according to claim 1, characterized in that: The time domain features are input into the gated recurrent unit for time modeling, and the frequency domain features are input into the squeeze-excited attention module for weighted processing to obtain weighted frequency domain features, including: inputting the time domain features into the gated recurrent unit, capturing the local time dependencies in the time domain features through CNN, and then capturing the long-term dependencies of the time domain features through the dynamic time modeling capability of GRU; the frequency domain features are input into the squeeze-excited attention module, firstly, global average pooling is performed on each feature channel in the frequency domain, global information is extracted and a vector describing the importance of the channel is generated, the vector is compressed and reconstructed to obtain the channel weight coefficient, and the frequency domain features are weighted using the weight coefficient.

6. The lithium-ion battery thermal runaway early warning method based on multi-modal reconstruction fusion according to claim 1, characterized in that: The cross-modal attention mechanism calculates the relationship between the weighted frequency domain features and the time domain features after time modeling, and dynamically weights the features in the frequency domain and time domain modes to achieve information fusion. The mathematical formula of the cross-modal attention mechanism is: Attention = softmax( QK T ) Fused_features = Attention× V Among them, the time domain features are Q and K , the frequency domain features are V , T is the transpose of matrix K.

7. A lithium-ion battery thermal runaway warning system based on multi-modal reconstruction fusion, characterized in that: include: A signal acquisition module is used to acquire a sound time domain signal of a lithium-ion battery and convert the sound time domain signal into a sound frequency domain signal by applying Fourier transform; A feature extraction module is used to determine the time delay and embedding dimension based on the phase space reconstruction method, perform high-dimensional embedding on the sound time domain signal and the sound frequency domain signal to reconstruct the high-dimensional phase space, and extract the phase space reconstruction features of the sound time domain signal and the sound frequency domain signal in the high-dimensional phase space to obtain the time domain features and the frequency domain features; The feature fusion module is used to input the time domain features into the gated recurrent unit for time modeling, and input the frequency domain features into the squeeze-excitation attention module for weighted processing to obtain weighted frequency domain features; the relationship between the weighted frequency domain features and the time domain features after time modeling is calculated through the cross-modal attention mechanism, and the weighted frequency domain features are dynamically fused with the time domain features after time modeling to obtain fused features; The classification module is used to input the fusion features into the fully connected layer, and output the classification results of thermal runaway of lithium-ion batteries through linear transformation and activation function.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the lithium-ion battery thermal runaway warning method based on multi-modal reconstruction fusion according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the lithium-ion battery thermal runaway warning method based on multi-modal reconstruction fusion as described in any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the lithium-ion battery thermal runaway warning method based on multi-modal reconstruction fusion as described in any one of claims 1 to 6.

Images