Lithium battery thermal runaway active protection method and system combined with self-supervised learning

Abstract

The invention discloses a lithium battery thermal runaway active protection method and system combined with self-supervised learning, and relates to the technical field of thermal runaway protection, and the method comprises the steps: detecting the dynamic impedance of a battery through a high-frequency excitation signal; learning data internal distribution characteristics by comparing loss functions; constructing a generator to simulate early-stage characteristics of thermal runaway, and optimizing the sensitivity of a simulation model to an abnormal mode through a discriminator; the early warning value is updated in real time to adapt to battery aging and working condition changes; establishing a multi-physics field coupling model; the simulation data and real experiment data are fused; dynamically adjusting the charging current based on the battery health state; and positioning a thermal runaway fault unit through the graph neural network. By arranging the abnormity simulation module, the self-supervised learning module and the model building module, abnormal signals in the early stage of thermal runaway are found in advance, active protection is achieved, battery aging and working condition changes are adapted, and the accuracy and reliability of early warning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal runaway protection, and specifically to an active protection method and system for lithium battery thermal runaway combined with self-supervised learning. Background Technique

[0002] With the continuous growth of the global demand for clean energy, lithium batteries have been widely used in fields such as electric vehicles, energy storage systems, and portable electronic devices due to their high energy density, long cycle life, and environmental friendliness. However, lithium batteries may experience thermal runaway during the charging and discharging process, which is an extremely dangerous situation that can lead to a sharp increase in battery temperature, gas release, and even fire and explosion, seriously threatening the safety of personnel and property.

[0003] Existing lithium battery thermal runaway protection technologies mainly rely on passive protection measures, such as setting safety valves and using flame retardant materials. Although these passive protection measures can reduce the harm of thermal runaway to a certain extent, they cannot fundamentally prevent the occurrence of thermal runaway. Moreover, as the battery ages and the operating conditions change, the performance of the battery will change, and the existing protection technologies are difficult to adapt to these changes, resulting in a decline in the protection effect. Summary of the Invention

[0004] To solve the above technical problems, an active protection method and system for lithium battery thermal runaway combined with self-supervised learning are provided, and the present technical solution solves the problems raised in the above background technique.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows: An active protection method for lithium battery thermal runaway combined with self-supervised learning, comprising: Monitoring the internal impedance phase shift, detecting the battery dynamic impedance through a high-frequency excitation signal, capturing the abnormal impedance phase shift in the early stage of thermal runaway, and analyzing the change in the characteristic gas concentration and the acoustic signal characteristics; Dividing the time series data of the normal charging and discharging process into "anchor point - positive sample - negative sample" triples, and learning the internal distribution characteristics of the data through a contrastive loss function; Constructing a generator to simulate the early characteristics of thermal runaway, and optimizing the sensitivity of the simulation model to abnormal patterns through a discriminator; Based on the statistical method of self-supervised learning, using exponential weighted moving average to update the warning value in real time to adapt to battery aging and operating condition changes; Establishing a multi-physics field coupling model to simulate the thermal runaway chain reaction process of lithium ion battery passivation film decomposition and negative electrode - electrolyte reaction, and generating training data; Fusing the simulation data with the real experimental data to improve the generalization ability of the model, quantifying the contribution degree of each input feature to the warning result, and optimizing the sensor deployment strategy; Dynamically adjust the charging current based on the battery health status, and control the battery temperature difference through a liquid cooling system; Combine the acoustic signal and the spatial distribution of gas concentration, locate the thermal runaway fault unit through a graph neural network, and trigger a local fuse.

[0006] Preferably, the construction of the generator simulates the early characteristics of thermal runaway, and the discriminator optimizes the sensitivity of the simulation model to abnormal patterns, specifically including: Obtain real early characteristic data through lithium battery thermal runaway experiments. The real early characteristic data includes the change sequences of voltage, current, temperature, characteristic gas concentration, and acoustic signal parameters over time; Collect the time-series data of the battery during normal charge and discharge, and compare it with the auxiliary generator to learn the time-series data in the normal mode; The generator adopts the generator structure in the generative adversarial network. The input is a random noise vector, and the output is a simulated sequence of early thermal runaway characteristic data; Initialize the network parameters of the generator. In each iteration, the generator receives a random noise vector as input and generates simulated early thermal runaway characteristic data; Input the generated simulated data and real early characteristic data into the discriminator for discrimination. Based on the feedback of the discriminator, use backpropagation to update the parameters of the generator; The discriminator adopts a combined structure based on a convolutional neural network and a fully connected neural network that matches the generator, extracts the features of the data, and classifies them; The input of the discriminator is the simulated data generated by the generator and the real early characteristic data, and the output is the probability that the input data is real data; Initialize the network parameters of the discriminator. In each iteration, input the generated simulated data and real early thermal runaway data into the discriminator respectively, and calculate the binary cross-entropy loss function value of the discriminator; Based on the value of the loss function, use backpropagation to update the parameters of the discriminator; The generator and the discriminator are alternately trained. In each iteration, first fix the parameters of the discriminator and train the generator, and then fix the parameters of the generator and train the discriminator.

[0007] Preferably, the self-supervised learning-based statistical method uses exponential weighted moving average to update the warning value in real time to adapt to battery aging and working condition changes, specifically including: Set a sliding window with a fixed length, and calculate the statistical features that reflect the changes of battery parameters within a local time range in each window. The statistical features include mean, variance, maximum value, and minimum value; Calculate the difference value of the battery parameters. The first-order difference value is obtained by subtracting the value at the previous moment from the value at the current moment, and the second-order difference is obtained by subtracting the previous first-order difference value from the current first-order difference value, which reflects the change trend and change rate of the battery parameters; Construct an autoencoder model, which consists of two parts: an encoder and a decoder; The encoder compresses the input feature vector into a low-dimensional latent space, and the decoder restores the vector in the latent space to the original feature vector; Use the unlabeled time-series feature data to train the autoencoder, minimize the reconstruction error between the input feature vector and the output feature vector of the decoder. Based on self-supervised learning, the autoencoder learns the internal structure and patterns in the data; For the temperature parameter of the battery, based on the normal operating temperature range of the battery, set an initial warning value higher than the upper limit of the normal operating temperature; For the battery parameter features after self-supervised learning feature extraction, use the exponentially weighted moving average formula to calculate the warning value at the current moment; Based on the normal operating range and safety requirements of the battery, set the anomaly detection threshold; Judge whether the real-time updated warning value exceeds the set anomaly detection threshold. If so, trigger an anomaly alarm to indicate that the battery has a risk of thermal runaway. If not, there is no output.

[0008] Preferably, the establishment of the multi-physics field coupling model to simulate the thermal runaway chain reaction process of the decomposition of the passivation film and the negative electrode-electrolyte reaction of the lithium-ion battery to generate training data specifically includes: Identify and construct the thermal field of the lithium battery thermal runaway based on the internal temperature distribution and change of the battery, identify and construct the chemical field of the lithium battery thermal runaway based on the chemical reaction process and product generation, and identify and construct the mechanical field of the lithium battery thermal runaway based on the battery structure deformation; Simplify the spatial model of the battery into a regular cylindrical shape and define the dimensions and positions of each part; Obtain the material properties of each component of the battery and assign the corresponding material properties to each part of the geometric model. The material properties include thermal conductivity, specific heat capacity, density, and chemical reaction kinetic parameters; Based on the analyzed physical fields, select the corresponding physical field interfaces. The physical field interfaces include heat transfer interfaces, chemical reaction engineering interfaces, and solid mechanics interfaces; Set boundary conditions for each physical field interface. In the heat transfer interface, set the heat dissipation condition on the battery surface. In the chemical reaction engineering interface, set the initial concentration of the reactants and the reaction rate expression; Set the coupling relationship between physical fields in the software. The heat released by the chemical reaction is coupled to the heat transfer interface as a heat source. The temperature change affects the chemical reaction rate, and a temperature-dependent reaction rate expression is introduced in the chemical reaction engineering interface; Start the model solving process to obtain the distribution and variation of various physical quantities of the battery during thermal runaway; Analyze the solving results and extract the curves of temperature, pressure, and substance concentration changing with time, as well as the distribution nephogram inside the battery.

[0009] Furthermore, a lithium battery thermal runaway active protection system combined with self-supervised learning is proposed to implement the lithium battery thermal runaway active protection method combined with self-supervised learning as described above, including: Anomaly monitoring module, which is used to monitor the internal impedance phase shift, detect the battery dynamic impedance through a high-frequency excitation signal, capture the impedance phase shift anomaly in the early stage of thermal runaway, and analyze the change of characteristic gas concentration and the characteristics of acoustic signals; Feature learning module, which is used to divide the time-series data of the normal charge and discharge process into "anchor point - positive sample - negative sample" triples, and learn the internal distribution characteristics of the data through a contrast loss function; Anomaly simulation module, which is used to construct a generator to simulate the early characteristics of thermal runaway, and optimize the sensitivity of the simulation model to abnormal patterns through a discriminator; Self-supervised learning module, which is used to update the warning value in real time to adapt to battery aging and working condition changes based on the statistical method of self-supervised learning and using exponential weighted moving average; Model construction module, which is used to establish a multi-physical field coupling model, simulate the thermal runaway chain reaction process of the decomposition of the lithium-ion battery passivation film and the negative electrode - electrolyte reaction, and generate training data; Strategy optimization module, which is used to fuse the simulation data with the real experimental data, improve the generalization ability of the model, quantify the contribution degree of each input feature to the warning result, and optimize the sensor deployment strategy; Active protection module, which is used to dynamically adjust the charging current based on the battery health state and control the battery temperature difference through a liquid cooling system; Runaway location module, which is used to combine the acoustic signal and the spatial distribution of gas concentration, locate the thermal runaway fault unit through a graph neural network, and trigger a local fuse.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This method is a statistical method based on self-supervised learning, which uses exponential weighted moving average to update the warning value in real time. The exponential weighted moving average can dynamically adjust the warning value according to new data, so that the warning value always matches the current state of the battery. Whether the battery is in the initial or later stage of aging, or under different charge and discharge conditions, this method can accurately issue warnings, improving the adaptability and stability of the protection system. By integrating simulation data with real experimental data, the model can learn richer features and patterns, enhancing the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 Flowchart of the active protection method for lithium battery thermal runaway combining self-supervised learning according to the present invention; Figure 2 Flowchart of the method for monitoring the internal impedance phase shift according to the present invention; Figure 3 Flowchart of the method for learning the intrinsic distribution characteristics of data through a contrast loss function according to the present invention; Figure 4 Flowchart of the method for optimizing the sensitivity of the simulation model to abnormal patterns by a discriminator according to the present invention; Figure 5 Flowchart of the method for real-time updating the warning value to adapt to battery aging and working condition changes according to the present invention; Figure 6 Flowchart of the method for establishing a multi-physical field coupling model according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0012] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations.

[0013] Referring to Figure 1 As shown, an active protection method for lithium battery thermal runaway combining self-supervised learning includes: Monitoring the internal impedance phase shift, detecting the dynamic impedance of the battery through a high-frequency excitation signal, capturing the impedance phase shift anomaly in the early stage of thermal runaway, and analyzing the changes in the concentration of characteristic gases and the characteristics of acoustic signals; Dividing the time-series data of the normal charge and discharge process into "anchor point - positive sample - negative sample" triples, and learning the intrinsic distribution characteristics of the data through a contrast loss function; Constructing a generator to simulate the early characteristics of thermal runaway, and optimizing the sensitivity of the simulation model to abnormal patterns through a discriminator; Based on a statistical method of self-supervised learning, using exponential weighted moving average to real-time update the warning value to adapt to battery aging and working condition changes; Build a multi-physics coupling model to simulate the thermal runaway chain reaction process of the decomposition of the passivation film and the negative electrode-electrolyte reaction in a lithium-ion battery, and generate training data; Fuse the simulation data with real experimental data to improve the generalization ability of the model, quantify the contribution of each input feature to the warning result, and optimize the sensor deployment strategy; Dynamically adjust the charging current based on the battery health state, and control the battery temperature difference through a liquid cooling system; Combine the acoustic signal with the spatial distribution of gas concentration, locate the thermal runaway fault cell through a graph neural network, and trigger a local fuse.

[0014] Refer to Figure 2 As shown, monitor the internal impedance phase shift, detect the battery dynamic impedance through a high-frequency excitation signal, capture the abnormal impedance phase shift in the early stage of thermal runaway, and analyze the changes in the characteristic gas concentration and the acoustic signal characteristics. Specifically include: Design and apply a high-frequency excitation signal to the lithium battery, use a sine wave form signal, and determine its frequency range based on the battery characteristics and research requirements; While applying the high-frequency excitation signal, use an impedance analyzer to measure the impedance value of the battery at each frequency, and calculate its phase shift as the phase difference between the imaginary part and the real part of the impedance; Continuously apply high-frequency excitation and impedance measurement to the battery to obtain the dynamic impedance data of the battery under at least one working condition, reflecting the changes in the internal electrochemical process and physical structure of the battery over time; Collect the impedance phase shift data of the battery under normal charge and discharge and static states, and establish a normal impedance phase shift model through statistical analysis; Compare the monitored impedance phase shift data with the normal impedance phase shift model in real time, and determine the abnormal degree of the impedance phase shift based on the deviation degree of the monitored impedance phase shift data from the normal model; Deploy gas sensors around the battery to monitor the characteristic gases generated during the thermal runaway process of the battery, and generate electrical signals through the chemical reaction between the gas and the sensor electrode to realize the measurement of the gas concentration; Install an acoustic sensor near the battery to collect the acoustic signals generated during the thermal runaway process of the battery, capture at least one frequency and intensity of the sound signal, and extract the acoustic signal characteristics, where the acoustic signal characteristics include frequency, amplitude, and time-domain waveform.

[0015] Apply a high - frequency excitation signal to the lithium - ion battery by combining a constant - current source or a constant - voltage source with a signal generator. The constant - current source can ensure the stability of the excitation current, while the constant - voltage source is suitable for scenarios with specific requirements for voltage excitation. Connect the output terminals of the signal generator to the positive and negative electrodes of the battery, and precisely control the characteristics of the excitation signal by adjusting the parameters of the signal generator such as amplitude, frequency, etc. When applying the signal, pay attention to avoiding damage to the battery caused by excessive signal amplitude. Generally, the signal amplitude is controlled between 1% - 5% of the rated voltage of the battery.

[0016] Refer to Figure 3 As shown, divide the timing data of the normal charge - discharge process into "anchor - positive sample - negative sample" triples, and learn the internal distribution characteristics of the data through the contrast loss function. Specifically include: Use the battery management system to continuously collect the change sequences of voltage, current, and temperature parameters of the lithium - ion battery over time during the normal charge - discharge process, and record them as timing data; Clean the collected original timing data to remove noise data and outliers; Select a data point as an anchor from the pre - processed timing data; Record the positive sample as the data point that belongs to the same normal charge - discharge process as the data point corresponding to the anchor and has the same characteristics; Record the negative sample as the data point that does not belong to the same normal charge - discharge process as the data point corresponding to the anchor and has different characteristics; Design a feature extraction network based on a convolutional neural network to map the data points in the triple to a feature space; Input the anchor, positive sample, and negative sample into the feature extraction network respectively to obtain their feature representations in the feature space. The input is the data point after standardization processing, and the output is a feature vector with a fixed dimension; Use the contrast loss function to reduce the distance between the anchor and the positive sample in the feature space and increase the distance between the anchor and the negative sample in the feature space.

[0017] The contrast loss function is: , In the formula, is the contrast loss function, is the anchor, is the positive sample, is the negative sample, are the Euclidean distances between the anchor and the positive sample, and the anchor and the negative sample in the feature space respectively, is the parameter that controls the distance interval between the positive sample and the negative sample; The selection of anchor points can be completely random or can be selected according to certain rules. One anchor point is selected every certain number of data points. Positive samples are recorded as data points corresponding to the data points of the anchor point that belong to the same normal charge-discharge process and have the same characteristics. Data points with similar voltage-current change trends in the same charge-discharge stage as the anchor point, such as the initial charging stage, the mid-charging stage, and the end-charging stage, can be selected as positive samples. Negative samples are recorded as data points corresponding to the data points of the anchor point that do not belong to the same normal charge-discharge process and have different characteristics. Data points with significantly different voltage-current change trends from the anchor point from different charge-discharge processes, such as the charge-discharge processes of different batteries or the charge-discharge processes of the same battery at different time periods, can be selected as negative samples.

[0018] Referring to Figure 4 As shown, a generator is constructed to simulate the early characteristics of thermal runaway, and the discriminator is used to optimize the sensitivity of the simulation model to abnormal patterns, specifically including: Real early characteristic data is obtained through lithium battery thermal runaway experiments. The real early characteristic data includes the change sequences of voltage, current, temperature, characteristic gas concentration, and acoustic signal parameters over time; The time-series data of the battery during normal charge-discharge processes is collected, and the auxiliary generator is compared to learn the time-series data in the normal mode; The generator adopts the generator structure in the generative adversarial network. The input quantity is a random noise vector, and the output quantity is a simulated early thermal runaway characteristic data sequence; The network parameters of the generator are initialized. In each iteration, the generator receives a random noise vector as input and generates simulated early thermal runaway characteristic data; The generated simulated data and the real early characteristic data are input into the discriminator for discrimination. Based on the feedback of the discriminator, the parameters of the generator are updated using backpropagation; The discriminator adopts a combined structure based on a convolutional neural network and a fully connected neural network that matches the generator to extract the features of the data and classify them; The input quantity of the discriminator is the simulated data generated by the generator and the real early characteristic data, and the output quantity is the probability that the input data is real data; The network parameters of the discriminator are initialized. In each iteration, the generated simulated data and the real early thermal runaway data are respectively input into the discriminator, and the binary cross-entropy loss function value of the discriminator is calculated; Based on the value of the loss function, the parameters of the discriminator are updated using backpropagation; The generator and the discriminator are alternately trained. In each iteration, first, the parameters of the discriminator are fixed to train the generator, and then the parameters of the generator are fixed to train the discriminator.

[0019] The backpropagation algorithm updates the discriminator parameters according to the gradient of the loss function with respect to the discriminator parameters using an optimization algorithm, enabling the discriminator to better distinguish between real data and simulated data, thereby enhancing the sensitivity to abnormal patterns. The generator and discriminator are alternately trained. In each iteration, the parameters of the discriminator are first fixed to train the generator so that the simulated data generated by the generator can deceive the discriminator as much as possible. Then, the parameters of the generator are fixed to train the discriminator so that the discriminator can more accurately distinguish between real data and simulated data. Appropriate training termination conditions are set, such as reaching a preset number of iterations, the value of the discriminator's loss function no longer decreasing significantly, or the performance of the generator and discriminator reaching equilibrium, etc. When the termination conditions are met, the training stops, and the trained generator and discriminator models are obtained.

[0020] Referring to Figure 5 As shown, a statistical method based on self-supervised learning uses exponential weighted moving average to update the warning value in real time to adapt to battery aging and operating conditions changes, which specifically includes: Set a sliding window with a fixed length, and calculate statistical features within each window that reflect the changes in battery parameters within a local time range. The statistical features include mean, variance, maximum value, and minimum value; Calculate the difference values of the battery parameters. Subtract the previous moment value from the current moment value to obtain the first-order difference value, and subtract the previous first-order difference value from the current first-order difference value to obtain the second-order difference, which reflects the change trend and change rate of the battery parameters; Construct an autoencoder model, which consists of two parts: an encoder and a decoder; The encoder compresses the input feature vector into a low-dimensional latent space, and the decoder restores the vector in the latent space to the original feature vector; Use unlabeled time-series feature data to train the autoencoder, minimizing the reconstruction error between the input feature vector and the output feature vector of the decoder. Based on self-supervised learning, the autoencoder learns the internal structure and patterns in the data; For the temperature parameter of the battery, based on the normal operating temperature range of the battery, set an initial warning value higher than the upper limit of the normal operating temperature; For the battery parameter features after self-supervised learning feature extraction, use the exponential weighted moving average formula to calculate the warning value at the current moment; Based on the normal operating range and safety requirements of the battery, set an abnormal detection threshold; Judge whether the warning value updated in real time exceeds the set abnormal detection threshold. If so, trigger an abnormal alarm to indicate that the battery has a risk of thermal runaway. If not, no output is made.

[0021] The exponential weighted moving average formula is: , where, is the warning value at the current moment, is the characteristic value at the current moment, is the warning value at the previous moment, is the smoothing factor, and its value range is between 0 and 1; Construct an autoencoder model, which consists of an encoder and a decoder. The encoder is usually composed of multiple layers of neural networks and is used to compress the input feature vector into a low-dimensional latent space. Assuming the input feature vector is 10-dimensional, the encoder can compress it into a 3-dimensional latent space. The decoder is also composed of multiple layers of neural networks and is used to restore the vector in the latent space to the original feature vector.

[0022] Refer to Figure 6 As shown, establish a multi-physics field coupling model to simulate the thermal runaway chain reaction process of the decomposition of the passivation film of the lithium-ion battery and the negative electrode-electrolyte reaction, and generate training data specifically including: Identify and construct the thermal field of the lithium battery thermal runaway based on the internal temperature distribution and change of the battery, identify and construct the chemical field of the lithium battery thermal runaway based on the chemical reaction process and product generation, and identify and construct the mechanical field of the lithium battery thermal runaway based on the battery structure deformation; Simplify the spatial model of the battery into a regular cylinder shape and define the dimensions and positions of each part; Obtain the material properties of each component of the battery and assign the corresponding material properties to each part of the geometric model. The material properties include thermal conductivity, specific heat capacity, density, and chemical reaction kinetic parameters; Based on the analyzed physical fields, select the corresponding physical field interfaces. The physical field interfaces include heat transfer interface, chemical reaction engineering interface, and solid mechanics interface; Set boundary conditions for each physical field interface. In the heat transfer interface, set the heat dissipation condition on the battery surface. In the chemical reaction engineering interface, set the initial concentration of the reactants and the reaction rate expression; Set the coupling relationship between each physical field in the software. The heat released by the chemical reaction is coupled to the heat transfer interface as a heat source. The temperature change affects the chemical reaction rate, and a temperature-related reaction rate expression is introduced in the chemical reaction engineering interface; Start the model solution process to obtain the distribution and change of various physical quantities of the battery during the thermal runaway process; Analyze the solution results and extract the curves of the changes of temperature, pressure, and substance concentration with time and the distribution nephogram inside the battery.

[0023] Visualize the physical quantities such as the solved temperature, pressure, and substance concentration to generate corresponding distribution contour maps and change curves. Through the temperature distribution contour map, the temperature changes at different positions inside the battery can be intuitively seen, and through the substance concentration change curve, the consumption of reactants and the generation process of products can be understood. Conduct statistical analysis on the solution results, calculate the average value, maximum value, minimum value, etc. of key physical quantities, and evaluate the danger level of the battery during thermal runaway.

[0024] Furthermore, based on the same inventive concept as the above-mentioned active protection method for lithium battery thermal runaway combined with self-supervised learning, this solution also proposes an active protection system for lithium battery thermal runaway combined with self-supervised learning, including: Anomaly monitoring module, which is used to monitor the internal impedance phase shift, detect the battery dynamic impedance through high-frequency excitation signals, capture the impedance phase shift anomaly in the early stage of thermal runaway, and analyze the change of characteristic gas concentration and acoustic signal characteristics; Feature learning module, which is used to divide the time-series data of the normal charge and discharge process into "anchor point - positive sample - negative sample" triples, and learn the internal distribution characteristics of the data through the contrast loss function; Anomaly simulation module, which is used to construct a generator to simulate the early characteristics of thermal runaway, and optimize the sensitivity of the simulation model to abnormal patterns through a discriminator; Self-supervised learning module, which is used to update the warning value in real time to adapt to battery aging and working condition changes based on the statistical method of self-supervised learning and using exponential weighted moving average; Model construction module, which is used to establish a multi-physical field coupling model to simulate the thermal runaway chain reaction process of the decomposition of the lithium-ion battery passivation film and the negative electrode - electrolyte reaction, and generate training data; Strategy optimization module, which is used to fuse the simulation data with the real experimental data, improve the model generalization ability, quantify the contribution degree of each input feature to the warning result, and optimize the sensor deployment strategy; Active protection module, which is used to dynamically adjust the charging current based on the battery health state and control the battery temperature difference through a liquid cooling system; Runaway location module, which is used to combine the acoustic signal with the spatial distribution of gas concentration, locate the thermal runaway fault unit through a graph neural network, and trigger a local fuse.

[0025] Still further, this solution also proposes a computer-readable storage medium, on which a computer-readable program is stored. When the computer-readable program is called, it executes the above-mentioned active protection method for lithium battery thermal runaway combined with self-supervised learning.

[0026] It is understandable that the storage medium can be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid state disk (SSD).

[0027] In summary, the advantages of the present invention are as follows: The method is a statistical method based on self-supervised learning, which uses exponential weighted moving average to update the warning value in real time. The exponential weighted moving average can dynamically adjust the warning value according to new data, so that the warning value always matches the current state of the battery. Whether the battery is in the initial stage or the later stage of aging, or under different charge and discharge conditions, this method can accurately issue a warning, improving the adaptability and stability of the protection system. By fusing simulation data with real experimental data, the model can learn richer features and patterns, improving the generalization ability of the model.

[0028] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An active protection method for thermal runaway of lithium batteries combined with self-supervised learning, characterized in that, Including: Monitoring the internal impedance phase shift, detecting the battery dynamic impedance through a high-frequency excitation signal, capturing the abnormal impedance phase shift in the early stage of thermal runaway, and analyzing the change of characteristic gas concentration and the characteristics of acoustic signals; Dividing the time-series data of the normal charge and discharge process into "anchor point - positive sample - negative sample" triples, and learning the internal distribution characteristics of the data through a contrastive loss function; Constructing a generator to simulate the characteristics in the early stage of thermal runaway, and optimizing the sensitivity of the simulation model to abnormal patterns through a discriminator; Based on the statistical method of self-supervised learning, using exponential weighted moving average to update the warning value in real time to adapt to battery aging and working condition changes; Establishing a multi-physics field coupling model to simulate the thermal runaway chain reaction process of the decomposition of the passivation film of lithium-ion batteries and the negative electrode - electrolyte reaction, and generating training data; Fusing the simulation data with the real experimental data to improve the generalization ability of the model, quantifying the contribution of each input feature to the warning result, and optimizing the sensor deployment strategy; Dynamically adjusting the charging current based on the battery health state, and controlling the battery temperature difference through a liquid cooling system; Combining the spatial distribution of acoustic signals and gas concentrations, locating the thermal runaway fault unit through a graph neural network, and triggering a local fuse.

2. The active protection method for thermal runaway of lithium batteries combining self-supervised learning according to claim 1, characterized in that The monitoring of the internal impedance phase shift, detecting the battery dynamic impedance through a high-frequency excitation signal, capturing the abnormal impedance phase shift in the early stage of thermal runaway, and analyzing the change of characteristic gas concentration and the characteristics of acoustic signals specifically include: Designing and applying a high-frequency excitation signal to the lithium battery, using a signal in the form of a sine wave, and determining its frequency range based on battery characteristics and research requirements; While applying the high-frequency excitation signal, using an impedance analyzer to measure the impedance value of the battery at each frequency, and calculating its phase shift as the phase difference between the imaginary part and the real part of the impedance; Continuously applying high-frequency excitation and impedance measurement to the battery to obtain the dynamic impedance data of the battery under at least one working condition, reflecting the changes of the internal electrochemical process and physical structure of the battery over time; Collecting the impedance phase shift data of the battery in the normal charge and discharge and static states, and establishing a normal impedance phase shift model through statistical analysis; Comparing the monitored impedance phase shift data with the normal impedance phase shift model in real time, and determining the abnormal degree of the impedance phase shift based on the deviation degree of the monitored impedance phase shift data from the normal model; Deploying gas sensors around the battery to monitor the characteristic gases generated during the thermal runaway process of the battery, and generating electrical signals through the chemical reaction between the gas and the sensor electrode to achieve the measurement of gas concentration; Installing acoustic sensors near the battery to collect the acoustic signals generated during the thermal runaway process of the battery, capturing at least one frequency and intensity of sound signals, and extracting the acoustic signal characteristics, where the acoustic signal characteristics include frequency, amplitude, and time-domain waveform.

3. The active protection method for thermal runaway of lithium batteries combining self-supervised learning according to claim 2, wherein The dividing the time-series data of the normal charge and discharge process into "anchor point - positive sample - negative sample" triples, and learning the internal distribution characteristics of the data through a contrastive loss function specifically includes: Using the battery management system to continuously collect the change sequences of voltage, current, and temperature parameters of the lithium battery during the normal charge and discharge process over time, and recording them as time-series data; Cleaning the collected original time-series data to remove noise data and outliers; Select a data point from the preprocessed time-series data as the anchor point; Record the positive samples as the data points that belong to the same normal charge-discharge process as the data point corresponding to the anchor point and have the same characteristics; Record the negative samples as the data points that do not belong to the same normal charge-discharge process as the data point corresponding to the anchor point and have different characteristics; Design a feature extraction network based on a convolutional neural network to map the data points in the triple to a feature space; Input the anchor point, positive samples, and negative samples into the feature extraction network respectively to obtain their feature representations in the feature space. The input is the data points after normalization processing, and the output is a feature vector with a fixed dimension; Use the contrastive loss function to reduce the distance between the anchor point and the positive samples in the feature space and increase the distance between the anchor point and the negative samples in the feature space.

4. A method for active protection against thermal runaway of a lithium battery combined with self-supervised learning according to claim 3, characterized in that The construction of the generator simulates the early characteristics of thermal runaway, and the discriminator optimizes the sensitivity of the simulation model to abnormal patterns, specifically including: Obtain real early characteristic data through lithium battery thermal runaway experiments. The real early characteristic data includes the change sequences of voltage, current, temperature, characteristic gas concentration, and acoustic signal parameters over time; Collect the time-series data of the battery during normal charge and discharge, and compare it to assist the generator in learning the time-series data in the normal mode; The generator adopts the generator structure in the generative adversarial network. The input is a random noise vector, and the output is a simulated early thermal runaway characteristic data sequence; Initialize the network parameters of the generator. In each iteration, the generator receives a random noise vector as input and generates simulated early thermal runaway characteristic data; Input the generated simulated data and the real early characteristic data into the discriminator for discrimination. Based on the feedback of the discriminator, use backpropagation to update the parameters of the generator; The discriminator adopts a combined structure based on a convolutional neural network and a fully connected neural network that matches the generator to extract the features of the data and classify them; The input of the discriminator is the simulated data generated by the generator and the real early characteristic data, and the output is the probability that the input data is real data; Initialize the network parameters of the discriminator. In each iteration, input the generated simulated data and the real early thermal runaway data into the discriminator respectively, and calculate the binary cross-entropy loss function value of the discriminator; Based on the value of the loss function, use backpropagation to update the parameters of the discriminator; The generator and the discriminator are alternately trained. In each iteration, first fix the parameters of the discriminator and train the generator, and then fix the parameters of the generator and train the discriminator.

5. The active protection method for thermal runaway of lithium batteries combining self-supervised learning according to claim 4, characterized in that The self-supervised learning-based statistical method uses exponential weighted moving average to update the warning value in real time to adapt to battery aging and operating conditions changes, specifically including: Set a sliding window with a fixed length, and calculate the statistical features that reflect the changes of battery parameters within a local time range within each window. The statistical features include mean, variance, maximum value, and minimum value; Calculate the difference values of the battery parameters. Subtract the previous moment value from the current moment value to obtain the first-order difference value, and subtract the previous first-order difference value from the current first-order difference value to obtain the second-order difference, which reflects the change trend and change rate of the battery parameters; Construct an autoencoder model, which consists of two parts: an encoder and a decoder; The encoder compresses the input feature vector into a low-dimensional latent space, and the decoder restores the vector in the latent space to the original feature vector; Use unlabeled time-series feature data to train the autoencoder, minimize the reconstruction error between the input feature vector and the output feature vector of the decoder, and based on self-supervised learning, the autoencoder learns the internal structure and patterns in the data; For the temperature parameter of the battery, based on the normal operating temperature range of the battery, set an initial warning value higher than the upper limit of the normal operating temperature; For the battery parameter features after self-supervised learning feature extraction, use the exponential weighted moving average formula to calculate the warning value at the current moment; Based on the normal operating range and safety requirements of the battery, set the anomaly detection threshold; Judge whether the real-time updated warning value exceeds the set anomaly detection threshold. If so, trigger an anomaly alarm to indicate that the battery has a risk of thermal runaway. If not, do not make an output.

6. The active protection method for thermal runaway of lithium batteries combined with self-supervised learning according to claim 5, wherein The establishment of the multi-physics coupling model to simulate the thermal runaway chain reaction process of the decomposition of the passivation film and the negative electrode-electrolyte reaction of the lithium-ion battery to generate training data specifically includes: Identify and construct the thermal field of the lithium battery thermal runaway based on the internal temperature distribution and changes of the battery, identify and construct the chemical field of the lithium battery thermal runaway based on the chemical reaction process and product generation, and identify and construct the mechanical field of the lithium battery thermal runaway based on the battery structure deformation; Simplify the spatial model of the battery into a regular cylindrical shape, and define the dimensions and positions of each part; Obtain the material properties of each component of the battery, and assign the corresponding material properties to each part of the geometric model. The material properties include thermal conductivity, specific heat capacity, density, and chemical reaction kinetic parameters; Based on the analyzed physical fields, select the corresponding physical field interfaces. The physical field interfaces include heat transfer interfaces, chemical reaction engineering interfaces, and solid mechanics interfaces; Set boundary conditions for each physical field interface. In the heat transfer interface, set the heat dissipation conditions on the battery surface. In the chemical reaction engineering interface, set the initial concentration of the reactants and the reaction rate expression; Set the coupling relationship between each physical field in the software. The heat released by the chemical reaction is coupled to the heat transfer interface as a heat source. The temperature change affects the chemical reaction rate, and a temperature-related reaction rate expression is introduced in the chemical reaction engineering interface; Start the model solving process to obtain the distribution and changes of various physical quantities of the battery during the thermal runaway process; Analyze the solution results, and extract the curves of the changes of temperature, pressure, and substance concentration over time and the distribution nephogram inside the battery.

7. A lithium battery thermal runaway active protection system combined with self-supervised learning is used to implement the lithium battery thermal runaway active protection method combined with self-supervised learning according to any one of claims 1-6, and is characterized in that, Including: Anomaly monitoring module, which is used to monitor the internal impedance phase shift, detect the battery dynamic impedance through high-frequency excitation signals, capture the impedance phase shift anomaly in the early stage of thermal runaway, and analyze the changes in the characteristic gas concentration and the acoustic signal characteristics; Feature learning module, which is used to divide the time-series data of the normal charge and discharge process into "anchor point-positive sample-negative sample" triples, and learn the internal distribution characteristics of the data through the contrast loss function; Anomaly simulation module, which is used to construct a generator to simulate the early characteristics of thermal runaway and optimize the sensitivity of the simulation model to abnormal patterns through a discriminator; Self-supervised learning module, which is used to update the early warning value in real time to adapt to battery aging and working condition changes by using exponential weighted moving average based on the statistical method of self-supervised learning; Model construction module, which is used to establish a multi-physics field coupling model, simulate the thermal runaway chain reaction process of the decomposition of the passivation film of lithium-ion batteries and the negative electrode-electrolyte reaction, and generate training data; Strategy optimization module, which is used to fuse simulation data with real experimental data, improve the generalization ability of the model, quantify the contribution of each input feature to the warning result, and optimize the sensor deployment strategy; Active protection module, which is used to dynamically adjust the charging current based on the battery health state and control the battery temperature difference through a liquid cooling system; Out-of-control positioning module, which is used to combine the acoustic signal and the spatial distribution of gas concentration, locate the thermal runaway fault unit through a graph neural network, and trigger a local fuse.

Images