Battery thermal runaway identification model training method and device, equipment, medium and program product

Abstract

The application relates to a battery thermal runaway identification model training method, device, equipment, medium and program product. The method comprises the following steps: acquiring a sound signal in a thermal runaway process of a battery, and cutting the sound signal to obtain various types of sound cut signals; determining a sound cut signal belonging to a safety valve opening sound type as a positive sample in a training sample, and determining a sound cut signal belonging to a noise type as a negative sample in the training sample; training a preliminary battery thermal runaway identification model through a training set in the positive sample and the negative sample; and updating the preliminary battery thermal runaway identification model through a verification set in the positive sample to obtain a target battery thermal runaway identification model. The method can improve the training accuracy of the battery thermal runaway identification model.

Description

Technical Field

[0001] This application relates to the field of big data technology, and in particular to a training method, apparatus, computer equipment, computer-readable storage medium, and computer program product for a battery thermal runaway identification model. Background Technology

[0002] Lithium-ion batteries are widely used in new energy vehicles, energy storage power stations, and portable electronic devices due to their high energy density and long cycle life. However, lithium batteries are prone to thermal runaway under conditions such as overcharging, over-discharging, short circuits, high temperatures, or mechanical impacts. Once thermal runaway occurs, the internal temperature of the battery rises rapidly in a short period of time, accompanied by electrolyte decomposition and violent gas expansion, ultimately leading to safety valve rupture, battery casing bulging, or even explosion, releasing toxic and harmful gases and causing fires, posing a serious threat to personal and property safety. In recent years, frequent accidents such as spontaneous combustion of new energy vehicles and fires in energy storage power stations have highlighted the urgent need for lithium battery thermal runaway early warning technology.

[0003] Current lithium battery thermal runaway early warning systems primarily rely on monitoring electrical parameters such as temperature, voltage, and internal resistance, as well as environmental monitoring methods like gas sensors and smoke sensors. The parameters obtained from these monitoring efforts are used to train a battery thermal runaway identification model. However, temperature monitoring suffers from significant lag; by the time a temperature sensor detects an abnormal temperature rise, the battery has often already entered the middle stages of thermal runaway, lacking sufficient emergency response time. Voltage and internal resistance monitoring are easily affected by battery consistency and aging, making it difficult to accurately capture early signs of thermal runaway. Gas and smoke sensors only respond when gas or smoke reaches their location and are susceptible to interference from ambient airflow, resulting in insufficient reliability of early warnings and low accuracy of the battery thermal runaway identification model. Therefore, a more accurate method for training battery thermal runaway identification models is urgently needed. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for training a battery thermal runaway identification model that can improve the training accuracy of the battery thermal runaway identification model, in order to address the above-mentioned technical problems.

[0005] In a first aspect, this application provides a training method for a battery thermal runaway identification model, the method comprising:

[0006] Acquire the sound signal during the thermal runaway process of the battery, and segment the sound signal to obtain various types of sound segmentation signals;

[0007] Sound segments belonging to the type of safety valve opening sound are identified as positive samples in the training samples, and sound segments belonging to the type of noise are identified as negative samples in the training samples.

[0008] A preliminary battery thermal runaway identification model is obtained by training using the training set from the positive samples and the negative samples.

[0009] The preliminary battery thermal runaway identification model is updated using the validation set in the positive samples to obtain the target battery thermal runaway identification model.

[0010] As one embodiment, the step of training a preliminary battery thermal runaway identification model using the training set from the positive samples and the negative samples includes:

[0011] Feature extraction is performed on the training set of the positive samples to obtain training features;

[0012] Based on the training features and the negative samples, construct the loss function of the Siamese neural network;

[0013] Based on the loss function, the twin neural network is trained to obtain a preliminary battery thermal runaway identification model.

[0014] As one embodiment, the step of constructing the loss function of the Siamese neural network based on the training features and the negative samples includes:

[0015] The training features are extracted using a Siamese neural network to obtain positive sample features; and the negative samples are extracted using a Siamese neural network to obtain negative sample features.

[0016] Obtain the temperature coefficient corresponding to the Siamese neural network, the positive sample similarity corresponding to the positive sample feature, and the negative sample similarity corresponding to the negative sample feature;

[0017] The loss function of the Siamese neural network is constructed based on the temperature coefficient, the positive sample similarity, and the negative sample similarity.

[0018] As one embodiment, obtaining the temperature coefficient corresponding to the Siamese neural network includes:

[0019] Obtain the model training requirements information corresponding to the target battery thermal runaway identification model;

[0020] The temperature coefficient corresponding to the Siamese neural network is constructed based on at least one of the model training requirement accuracy and model training requirement duration represented by the model training requirement information.

[0021] As one embodiment, the step of extracting features from the training set in the positive samples to obtain training features includes:

[0022] Short-time energy features are extracted from the training set in the positive samples to obtain short-time energy features.

[0023] The zero-crossing rate feature is obtained by extracting the number of times the training set in the positive samples crosses the zero level per unit time.

[0024] The training set in the positive samples is converted from the time domain to the Mel spectral domain to obtain the spectral envelope features;

[0025] The training features are obtained by fusing the short-time energy features, the zero-crossing rate features, and the spectral envelope features.

[0026] As one embodiment, after updating the preliminary battery thermal runaway identification model using the validation set from the positive samples to obtain the target battery thermal runaway identification model, the method further includes:

[0027] Obtain the thermal runaway identification result output by the target battery thermal runaway identification model;

[0028] If the probability of thermal runaway as represented by the thermal runaway identification result is greater than a first preset probability threshold and less than a second preset probability threshold, output thermal runaway warning information.

[0029] If the probability of thermal runaway as represented by the thermal runaway identification result is greater than the second preset probability threshold, the charging and discharging circuit of the battery corresponding to the thermal runaway identification result is cut off, the fire protection system corresponding to the thermal runaway identification result is controlled to be turned on, and thermal runaway alarm information is output.

[0030] Secondly, this application also provides a training device for a battery thermal runaway identification model, comprising:

[0031] The acquisition module is used to acquire the sound signals during the thermal runaway process of the battery;

[0032] The sample segmentation module is used to segment the sound signal to obtain various types of sound segmentation signals. The sound segmentation signals belonging to the safety valve opening sound type are determined as positive samples in the training samples, and the sound segmentation signals belonging to the noise type are determined as negative samples in the training samples.

[0033] The training module is used to train a preliminary battery thermal runaway identification model using the training set in the positive samples and the negative samples; and to update the preliminary battery thermal runaway identification model using the validation set in the positive samples to obtain a target battery thermal runaway identification model.

[0034] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the processor performs the following steps when executing the computer program:

[0035] Acquire the sound signal during the thermal runaway process of the battery, and segment the sound signal to obtain various types of sound segmentation signals;

[0036] Sound segments belonging to the type of safety valve opening sound are identified as positive samples in the training samples, and sound segments belonging to the type of noise are identified as negative samples in the training samples.

[0037] A preliminary battery thermal runaway identification model is obtained by training using the training set from the positive samples and the negative samples.

[0038] The preliminary battery thermal runaway identification model is updated using the validation set in the positive samples to obtain the target battery thermal runaway identification model.

[0039] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0040] Acquire the sound signal during the thermal runaway process of the battery, and segment the sound signal to obtain various types of sound segmentation signals;

[0041] Sound segments belonging to the type of safety valve opening sound are identified as positive samples in the training samples, and sound segments belonging to the type of noise are identified as negative samples in the training samples.

[0042] A preliminary battery thermal runaway identification model is obtained by training using the training set from the positive samples and the negative samples.

[0043] The preliminary battery thermal runaway identification model is updated using the validation set in the positive samples to obtain the target battery thermal runaway identification model.

[0044] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0045] Acquire the sound signal during the thermal runaway process of the battery, and segment the sound signal to obtain various types of sound segmentation signals;

[0046] Sound segments belonging to the type of safety valve opening sound are identified as positive samples in the training samples, and sound segments belonging to the type of noise are identified as negative samples in the training samples.

[0047] A preliminary battery thermal runaway identification model is obtained by training using the training set from the positive samples and the negative samples.

[0048] The preliminary battery thermal runaway identification model is updated using the validation set in the positive samples to obtain the target battery thermal runaway identification model.

[0049] The training method, apparatus, computer equipment, computer-readable storage medium, and computer program product of the aforementioned battery thermal runaway identification model acquire sound signals during the battery thermal runaway process and segment the sound signals to obtain various types of sound segmentation signals; the sound segmentation signals belonging to the safety valve opening sound type are determined as positive samples in the training samples, and the sound segmentation signals belonging to the noise type are determined as negative samples in the training samples; a preliminary battery thermal runaway identification model is obtained by training with the training set of the positive samples and the negative samples; the preliminary battery thermal runaway identification model is updated with the validation set of the positive samples to obtain the target battery thermal runaway identification model.

[0050] In this way, by segmenting the complex sound signals during the thermal runaway of the battery, it is possible to accurately identify the sound signals that belong to the safety valve opening sound type and the sound signals that belong to the noise type. By identifying the sound signals that belong to the safety valve opening sound type and the sound signals that belong to the noise type as positive samples and negative samples respectively, the comparative learning training of the battery thermal runaway identification model can be realized, thereby improving the training accuracy of the battery thermal runaway identification model. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is an application environment diagram of the training method for the battery thermal runaway identification model in one embodiment;

[0053] Figure 2 This is a flowchart illustrating the training method of a battery thermal runaway identification model in one embodiment;

[0054] Figure 3 This is a flowchart illustrating the steps of training a preliminary battery thermal runaway identification model using the training set in the positive samples and the negative samples in one embodiment.

[0055] Figure 4 This is a structural block diagram of a training device for a battery thermal runaway identification model in one embodiment;

[0056] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0059] It should be noted that this method employs multi-scale entropy features and backpropagation neural networks to identify thermal runaway phenomena in batteries, and utilizes wavelet transform and genetic algorithms for noise reduction of the acoustic signals. However, this method primarily relies on traditional artificial features and shallow neural networks, resulting in limited adaptability to complex noise environments and difficulty in handling the variable and highly interfering sound fields within energy storage power stations and battery compartments. Furthermore, fixed-rule denoising methods such as wavelets and genetic algorithms lack robustness and cannot effectively preserve the high-frequency transient features in the explosion sound of safety valves, ultimately affecting the stability of the identification. Moreover, this scheme only provides binary classification warning results, lacking refined risk grading and the ability to integrate with backend systems.

[0060] It should be noted that the scheme extracts features using time-frequency transformation and phase space reconstruction, and processes these features through a GRU (Gated Recurrent Unit) and an attention mechanism to ultimately achieve sound signal recognition. This approach is complex, computationally intensive, and difficult to deploy in real-time within battery management systems or field edge devices. Furthermore, it does not optimize features for the critical explosive transient acoustic event of a safety valve opening; general time-frequency and phase space features are insufficient to accurately capture the high-frequency, short-duration characteristics of the safety valve opening sound, resulting in inadequate targeting for early thermal runaway identification. More importantly, the scheme lacks a robust risk level classification mechanism, failing to meet the engineering requirements of energy storage systems for rapid response and automatic linkage.

[0061] Based on this, in order to overcome the technical defects of the aforementioned battery thermal runaway identification schemes, as an embodiment, the training method of the battery thermal runaway identification model provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, the sound acquisition component 102 communicates with the server 104 via a network. The sound acquisition component 102 is used to acquire sound signals during the thermal runaway process of the battery. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104 or placed on the cloud or other network servers. The server 104 acquires the sound signals during the thermal runaway process of the battery acquired by the sound acquisition component 102, and segments the sound signals to obtain various types of sound segmented signals; the sound segmented signals belonging to the safety valve opening sound type are determined as positive samples in the training samples, and the sound segmented signals belonging to the noise type are determined as negative samples in the training samples; a preliminary battery thermal runaway identification model is obtained by training with the training set and negative samples in the positive samples; the preliminary battery thermal runaway identification model is updated with the validation set in the positive samples to obtain the target battery thermal runaway identification model. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The sound acquisition component 102 can be a device with the target model deployed.

[0062] In one exemplary embodiment, such as Figure 2 As shown, a training method for a battery thermal runaway identification model is provided, which can be applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 202 to 208. Wherein:

[0063] Step 202: Acquire the sound signal during the thermal runaway process of the battery, and segment the sound signal to obtain various types of sound segmentation signals.

[0064] In step 202, the sound signal is acquired by a sound acquisition component, which can be a microphone. The microphone's sampling frequency is not lower than 48kHz, and its frequency response range should cover 100Hz-30kHz.

[0065] Therefore, considering that the frequency range of the safety valve opening sound signal and noise signal is 100Hz-30kHz, it is possible to accurately acquire the safety valve opening sound signal and noise signal. By ensuring that the microphone sampling frequency is not too low, aliasing and distortion of high-frequency sound signals in the acquired sound signal are avoided, which would lead to waveform distortion and the inability to identify the individual sound segmentation signals.

[0066] For example, acquiring sound signals during the thermal runaway process of a battery includes: acquiring sound signals during the thermal runaway process of a battery collected by a sound acquisition component.

[0067] As one embodiment, the sound signal is segmented to obtain various types of sound segmentation signals, including: filtering the sound signal to obtain high-frequency signals and low-frequency signals; identifying the high-frequency signals as signals belonging to the safety valve opening sound type and the low-frequency signals as signals belonging to the noise type; segmenting the signals belonging to the safety valve opening sound type according to a preset sound duration to obtain sound segmentation signals belonging to the safety valve opening sound type; and segmenting the signals belonging to the noise type according to a preset sound duration to obtain sound segmentation signals belonging to the noise type.

[0068] The preset sound duration can be an empirical value, such as 3 seconds.

[0069] As one embodiment, filtering the sound signal to obtain high-frequency and low-frequency signals includes: filtering the sound signal through a high-frequency bandpass filter to obtain a high-frequency signal, and identifying signals other than high-frequency signals in the sound signal as low-frequency signals.

[0070] The passband range of a high-frequency bandpass filter can be an empirical value, such as 6kHz-25kHz. The passband range of a high-frequency bandpass filter can be achieved by using an FIR window function (a finite impulse response (FIR) filter, which truncates the ideal infinite impulse response into a finite-length weighted function to suppress spectral leakage and Gibbs oscillation caused by truncation).

[0071] Optionally, the above method further includes: performing noise reduction processing on the high-frequency signal.

[0072] The noise reduction method can be achieved using the LMS (Least Mean Square) algorithm.

[0073] Step 204: The sound segmentation signal belonging to the safety valve opening sound type is determined as a positive sample in the training samples, and the sound segmentation signal belonging to the noise type is determined as a negative sample in the training samples.

[0074] In step 204, the positive samples include the training set and the validation set.

[0075] As one embodiment, determining the sound segmentation signal belonging to the safety valve opening sound type as a positive sample in the training samples includes: dividing the sound segmentation signal belonging to the safety valve opening sound type into a training set and a validation set of positive samples in the training samples according to a preset segmentation ratio.

[0076] The preset segmentation ratio can be an empirical value, such as 7:3, 8:2, 6:4, etc., and there are no restrictions here.

[0077] Step 206: Train the preliminary battery thermal runaway identification model using the training set from the positive samples and the negative samples.

[0078] Among them, the preliminary battery thermal runaway identification model in step 206 is a twin neural network model.

[0079] For example, step 206 includes: training a preliminary battery thermal runaway identification model by contrastive learning according to the training set in the positive samples and the negative samples.

[0080] Step 208: Update the preliminary battery thermal runaway identification model using the validation set in the positive samples to obtain the target battery thermal runaway identification model.

[0081] For example, step 208 includes: the validation set includes multiple first samples, each first sample includes a first input signal and a first real label corresponding to the first input signal; for each first sample, a first output result is obtained by inputting the first input signal of the first sample into the preliminary battery thermal runaway identification model; a first loss of the preliminary battery thermal runaway identification model is constructed based on the difference between the first input result and the first real label; if the first loss table converges, the current preliminary battery thermal runaway identification model is determined as the target battery thermal runaway identification model; if the first loss table does not converge, the preliminary battery thermal runaway identification model is updated, and the process returns to the step of obtaining the first output result by inputting the first input signal of the first sample into the preliminary battery thermal runaway identification model, until the first loss table converges.

[0082] The methods for updating the model include, but are not limited to, gradient ascent and gradient descent.

[0083] Optionally, after step 208, the training method for the battery thermal runaway identification model further includes: obtaining the thermal runaway identification result output by the target battery thermal runaway identification model; outputting thermal runaway warning information when the probability of thermal runaway occurrence represented by the thermal runaway identification result is greater than a first preset probability threshold and less than a second preset probability threshold; and cutting off the charging and discharging circuit of the battery corresponding to the thermal runaway identification result, controlling the fire protection system corresponding to the thermal runaway identification result to open, and outputting thermal runaway alarm information when the probability of thermal runaway occurrence represented by the thermal runaway identification result is greater than the second preset probability threshold.

[0084] Thus, when the probability of thermal runaway is greater than the first preset probability threshold but less than the second preset probability threshold, meaning that the battery corresponding to the thermal runaway identification result has a certain risk of thermal runaway, a thermal runaway warning message is output to notify relevant personnel to intervene in a timely manner. When the probability of thermal runaway is greater than the second preset probability threshold, meaning that the battery corresponding to the thermal runaway identification result has a high risk of thermal runaway, the charging and discharging circuit is cut off in time to prevent the battery from continuing to heat up, and the fire suppression system is activated to extinguish the fire in time before the battery temperature becomes too high and causes a fire.

[0085] The first preset probability threshold and the second preset probability threshold can be empirical values. For example, the first preset probability threshold can be 60% and the second preset probability threshold can be 80%.

[0086] As one embodiment, obtaining the thermal runaway identification result output by the target battery thermal runaway identification model includes: acquiring the target sound signal of the target battery, inputting the target sound signal into the target battery thermal runaway identification model, and obtaining the thermal runaway identification result output by the target battery thermal runaway identification model based on the target sound signal.

[0087] Among them, thermal runaway early warning information has a higher level of alertness than thermal runaway alarm information. For example, thermal runaway early warning information may only include text information, while thermal runaway alarm information may include both text information and alarm sound.

[0088] The probability of thermal runaway can include at least one of the probability of suspected thermal runaway and the probability of confirmed thermal runaway.

[0089] As one embodiment, cutting off the charging and discharging circuit of the battery corresponding to the thermal runaway identification result includes: sending a circuit cut-off command to the battery management system of the battery corresponding to the thermal runaway identification result, and cutting off the charging and discharging circuit of the battery corresponding to the thermal runaway identification result by the battery management system according to the circuit cut-off command.

[0090] In this way, by promptly cutting off the charging and discharging circuit of the battery corresponding to the thermal runaway identification result, the continuous heating of the battery corresponding to the thermal runaway identification result can be prevented, thus avoiding phenomena such as fire.

[0091] In the training method of the aforementioned battery thermal runaway identification model, sound signals during the battery's thermal runaway process are acquired and segmented into various types of sound segments. Sound segments belonging to the safety valve opening sound type are designated as positive samples in the training samples, while sound segments belonging to the noise type are designated as negative samples. A preliminary battery thermal runaway identification model is obtained through training with the positive and negative samples. The preliminary model is then updated using the validation set of the positive samples to obtain the target battery thermal runaway identification model. Thus, by segmenting the complex sound signals during battery thermal runaway, it is possible to accurately identify sound signals belonging to the safety valve opening sound type and those belonging to the noise type. By designating these as positive and negative samples respectively, comparative learning training of the battery thermal runaway identification model can be achieved, improving the training accuracy of the model.

[0092] In one exemplary embodiment, such as Figure 3 As shown, step 206 includes steps 302 to 306. Wherein:

[0093] Step 302: Extract features from the training set in the positive samples to obtain training features.

[0094] In step 302, the training set includes multiple second samples, each of which includes a second input signal and a second true label corresponding to the second input signal.

[0095] For example, step 302 includes: extracting short-time energy features from the training set in the positive samples to obtain short-time energy features; extracting the number of times the training set in the positive samples crosses the zero level per unit time to obtain zero-crossing rate features; converting the training set in the positive samples from the time domain to the Mel spectrum domain to obtain spectral envelope features; and fusing the short-time energy features, zero-crossing rate features, and spectral envelope features to obtain training features.

[0096] As an example, short-time energy feature extraction is performed on the training set in the positive samples to obtain short-time energy features, including: for the second input signal in each second sample in the training set in the positive samples, the second input signal is divided into multiple short frames, and the signal energy of each short frame is determined frame by frame to obtain short-time energy features.

[0097] Therefore, considering that the short-term energy of the sound of a safety valve opening will rise rapidly during a sudden signal, forming a significant energy peak, which is significantly different from the steady energy change of the background noise, the precise feature extraction of the sound of a safety valve opening can be achieved by extracting the short-term energy characteristics.

[0098] As an example, the zero-crossing rate feature is obtained by extracting the number of times the training set in the positive samples crosses the zero level per unit time. This includes: for the second input signal in each second sample in the training set of the positive samples, determining the number of times the second input signal crosses the zero level per unit time to obtain the zero-crossing rate feature.

[0099] Thus, considering that background noise is mostly low-frequency clutter with a low and stable zero-crossing rate, while the sound of a safety valve opening has many high-frequency components (explosive and ejection), and its waveform frequently crosses zero, resulting in a sudden increase in the zero-crossing rate, accurate feature extraction of the safety valve opening sound can be achieved by extracting the zero-crossing rate characteristics.

[0100] As one embodiment, converting the training set in the positive samples from the time domain to the Mel spectrum domain to obtain spectral envelope features includes: for the second input signal in each second sample in the training set of the positive samples, converting the second input signal from the time domain to the Mel spectrum domain to obtain spectral envelope features.

[0101] Thus, since the Mel spectrum domain can well reflect the spectral envelope characteristics of the sound signal, and the spectral envelope characteristics have a strong characterization ability for the unique high-frequency spectral characteristics of the safety valve opening sound, the accurate feature extraction of the safety valve opening sound can be achieved by extracting the spectral envelope characteristics.

[0102] As one embodiment, the training features are obtained by fusing short-time energy features, zero-crossing rate features and spectral envelope features, including: splicing short-time energy features, zero-crossing rate features and spectral envelope features to obtain training features.

[0103] Thus, by comprehensively utilizing the short-time energy characteristics, zero-crossing rate characteristics, and spectral envelope characteristics obtained through the above three methods, the characteristics of the safety valve opening sound can be extracted comprehensively and accurately, providing rich and effective data features for subsequent battery thermal runaway identification models.

[0104] As another embodiment, the training features are obtained by fusing short-time energy features, zero-crossing rate features and spectral envelope features, including: weighted fusing of short-time energy features, zero-crossing rate features and spectral envelope features to obtain training features.

[0105] Thus, by weighting and fusing the short-time energy features, zero-crossing rate features, and spectral envelope features obtained through the above three methods, the importance of the short-time energy features, zero-crossing rate features, and spectral envelope features for the feature extraction of the safety valve opening sound can be quantified by the weights, thereby enabling the short-time energy features, zero-crossing rate features, and spectral envelope features to provide corresponding contribution ratios to the training features.

[0106] Thus, the training set, as the core data source for model feature learning and acoustic pattern mining, with its large proportion of positive samples, is sufficient to fully mine and characterize the common acoustic features of battery thermal runaway acoustic signals. After extracting feature parameters through short-time energy, zero-crossing rate, and Mel frequency cepstral coefficients, a stable and generalizable thermal runaway acoustic feature system can be constructed, providing a benchmark for subsequent model training and feature matching. The core role of the validation set is to serve as an independent sample set, objectively verifying the effectiveness and reliability of the acoustic features mined from the training set and evaluating their generalization ability. If the same feature extraction process is performed on the validation set simultaneously in advance, it will cause premature leakage of feature information, lose the independent verification attribute of the validation set, and make it impossible to truly verify the generalization ability and recognition accuracy of the extracted thermal runaway acoustic features. It will also be difficult to achieve error verification, vulnerability investigation, and optimization of the model feature system, violating the basic experimental principle of isolating training and validation datasets in machine learning and acoustic feature modeling. Therefore, this case only performs the aforementioned short-time energy features, zero-crossing rate features, and spectral envelope features on the training set, and does not perform the aforementioned short-time energy features, zero-crossing rate features, and spectral envelope features on the validation set.

[0107] Step 304: Construct the loss function of the Siamese neural network based on the training features and negative samples.

[0108] For example, step 304 includes: extracting features from the training set in the positive samples to obtain training features; constructing a loss function for the Siamese neural network based on the training features and negative samples; and training the Siamese neural network based on the loss function to obtain a preliminary battery thermal runaway identification model.

[0109] As one embodiment, a loss function for the Siamese neural network is constructed based on training features and negative samples, including: extracting features from the training features using the Siamese neural network to obtain positive sample features; and extracting features from the negative samples using the Siamese neural network to obtain negative sample features; obtaining the temperature coefficient corresponding to the Siamese neural network, the positive sample similarity corresponding to the positive sample features, and the negative sample similarity corresponding to the negative sample features; and constructing the loss function for the Siamese neural network based on the temperature coefficient, the positive sample similarity, and the negative sample similarity.

[0110] The twin neural network adopts a two-branch backbone structure, and the weights of the two branches are completely shared.

[0111] As one embodiment, obtaining the temperature coefficient corresponding to the Siamese neural network includes: obtaining model training requirement information corresponding to the target battery thermal runaway identification model; and constructing the temperature coefficient corresponding to the Siamese neural network based on at least one of the model requirement accuracy and model training requirement duration represented by the model training requirement information.

[0112] Among them, the model training requirement information can be empirical values ​​or set by the user as needed; the temperature coefficient is inversely correlated with the model requirement accuracy represented by the model training requirement information, and the temperature coefficient is inversely correlated with the model training requirement duration represented by the model training requirement information.

[0113] Thus, a smaller temperature parameter strengthens the difference constraint in the feature embedding of positive and negative samples, increasing the difficulty for the model to distinguish them; a larger temperature parameter smooths the similarity probability distribution, weakens the interference of difficult negative samples, ensures gradient stability and better convergence during model training, and adapts to the preset cosine similarity threshold constraint for positive and negative samples. Therefore, setting the temperature coefficient to be inversely correlated with the model accuracy requirement represented by the model training requirement information, and setting the temperature coefficient to be inversely correlated with the model training duration requirement represented by the model training requirement information, improves the accuracy of temperature coefficient generation.

[0114] As one embodiment, the loss function of the Siamese neural network is constructed based on the temperature coefficient, positive sample similarity, and negative sample similarity, including: constructing the negative logarithm of the positive sample matching probability based on the temperature coefficient, positive sample similarity, and negative sample similarity to obtain the loss function of the Siamese neural network.

[0115] Furthermore, based on the temperature coefficient, positive sample similarity, and negative sample similarity, a positive sample matching probability is constructed, including: based on the temperature coefficient, positive sample similarity, and negative sample similarity, constructing the negative logarithm of the positive sample matching probability, determining the positive sample selection probability corresponding to the candidate set consisting of 1 positive sample and K negative samples, and determining the positive sample weight probability as the positive sample matching probability.

[0116] Optionally, based on the temperature coefficient, positive sample similarity, and negative sample similarity, the negative logarithm of the positive sample matching probability is constructed, and the loss function of the Siamese neural network can be expressed by the following formula:

[0117]

[0118] in, The loss value for the Siamese neural network. For positive sample similarity, For negative sample similarity, This is the temperature coefficient.

[0119] Step 306: Based on the loss function, train the Siamese neural network to obtain a preliminary battery thermal runaway identification model.

[0120] For example, step 306 includes: determining the model loss value of the Siamese neural network based on the loss function, solving the gradient based on the model loss value, updating the shared weight parameters of the Siamese neural network based on the gradient; returning to the step of constructing the loss function of the Siamese neural network based on the training features and negative samples, minimizing the model loss value as the optimization objective, and obtaining a preliminary battery thermal runaway identification model.

[0121] In this embodiment, training features are obtained by extracting features from the training set of positive samples; a loss function for the Siamese neural network is constructed based on the training features and negative samples; the Siamese neural network is trained according to the loss function to obtain a preliminary battery thermal runaway identification model. By extracting effective features from positive samples and constructing loss constraints using negative samples, the Siamese network learns the specific features of thermal runaway sounds under the optimization of the loss function. This reduces the feature distance between similar thermal runaway samples and increases the feature distance between samples and background noise in the feature space, improving feature discrimination. A positive and negative sample comparison training scheme is adopted. For single-classification samples requiring massive amounts of finely labeled data, network iterative optimization can be completed solely through sample pairing, effectively reducing manual labeling costs and adapting to the actual working conditions where thermal runaway audio samples are scarce in industrial scenarios. By relying on the differential constraints of the loss function on positive and negative samples, the model can weaken the interference caused by environmental noise and background noise in industrial sites, improve the robustness of the initial recognition model under complex working conditions, reduce misjudgments caused by non-target noise, and ensure the accuracy of subsequent battery thermal runaway recognition. Through the structure of shared weights in the Siamese network, the direction of comparative loss optimization is clear, the network converges faster, and the initial recognition model can be constructed efficiently.

[0122] As a detailed embodiment, the sound signal during the thermal runaway of the battery is acquired, and the sound signal is segmented to obtain various types of sound segmentation signals; the sound segmentation signal belonging to the safety valve opening sound type is determined as the positive sample in the training sample, and the sound segmentation signal belonging to the noise type is determined as the negative sample in the training sample; short-time energy features are extracted from the training set in the positive samples to obtain short-time energy features; the number of times the training set in the positive samples crosses the zero level per unit time is extracted to obtain the zero-crossing rate feature; the training set in the positive samples is converted from the time domain to the Mel spectrum domain to obtain the spectral envelope feature; the short-time energy feature, the zero-crossing rate feature, and the spectral envelope feature are fused to obtain the training features.

[0123] Furthermore, feature extraction is performed on the training features using a Siamese neural network to obtain positive sample features; and feature extraction is performed on the negative samples using a Siamese neural network to obtain negative sample features; model training requirement information corresponding to the target battery thermal runaway identification model is obtained; based on at least one of the model training requirement accuracy and model training requirement duration represented by the model training requirement information, a temperature coefficient corresponding to the Siamese neural network is constructed, and the positive sample similarity corresponding to the positive sample features and the negative sample similarity corresponding to the negative sample features are obtained; based on the temperature coefficient, positive sample similarity, and negative sample similarity, a loss function for the Siamese neural network is constructed; based on the loss function, the Siamese neural network is... A preliminary battery thermal runaway identification model is obtained by training the network; the preliminary battery thermal runaway identification model is updated using the validation set in the positive samples to obtain the target battery thermal runaway identification model; the thermal runaway identification result output by the target battery thermal runaway identification model is obtained; when the probability of thermal runaway occurrence represented by the thermal runaway identification result is greater than a first preset probability threshold and less than a second preset probability threshold, thermal runaway warning information is output; when the probability of thermal runaway occurrence represented by the thermal runaway identification result is greater than the second preset probability threshold, the charging and discharging circuit of the battery corresponding to the thermal runaway identification result is cut off, the fire protection system corresponding to the thermal runaway identification result is activated, and thermal runaway alarm information is output.

[0124] Thus, by acquiring the sound signals during battery thermal runaway and segmenting them into various types, sound segments belonging to the safety valve opening sound type are identified as positive samples in the training samples, while sound segments belonging to the noise type are identified as negative samples. A preliminary battery thermal runaway identification model is obtained by training with the training set of positive samples and the negative samples. The preliminary model is then updated with the validation set of positive samples to obtain the target battery thermal runaway identification model. By segmenting the complex sound signals during battery thermal runaway, it is possible to accurately identify sound signals belonging to the safety valve opening sound type and those belonging to the noise type. Identifying these as positive and negative samples respectively allows for comparative learning training of the battery thermal runaway identification model, improving its training accuracy.

[0125] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0126] Based on the same inventive concept, this application also provides a training device for a battery thermal runaway identification model, which is used to implement the training method for the battery thermal runaway identification model described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the battery thermal runaway identification model training device provided below can be found in the limitations of the battery thermal runaway identification model training method described above, and will not be repeated here.

[0127] In one exemplary embodiment, such as Figure 4 As shown, a training device for a battery thermal runaway identification model is provided, comprising: an acquisition module, a sample partitioning module, and a training module, wherein:

[0128] The acquisition module is used to acquire the sound signals during the thermal runaway process of the battery.

[0129] The sample segmentation module is used to segment the sound signal to obtain various types of sound segmentation signals. Sound segmentation signals belonging to the safety valve opening sound type are identified as positive samples in the training samples, and sound segmentation signals belonging to the noise type are identified as negative samples in the training samples.

[0130] The training module is used to train a preliminary battery thermal runaway identification model using the training set and negative samples from the positive samples; and to update the preliminary battery thermal runaway identification model using the validation set from the positive samples to obtain the target battery thermal runaway identification model.

[0131] In one embodiment, the training module is further configured to: extract features from the training set in the positive samples to obtain training features; construct a loss function for the Siamese neural network based on the training features and negative samples; and train the Siamese neural network based on the loss function to obtain a preliminary battery thermal runaway identification model.

[0132] In one embodiment, the training module is further configured to: extract features from training features using a Siamese neural network to obtain positive sample features; and extract features from negative samples using a Siamese neural network to obtain negative sample features; obtain the temperature coefficient corresponding to the Siamese neural network, the positive sample similarity corresponding to the positive sample features, and the negative sample similarity corresponding to the negative sample features; and construct the loss function of the Siamese neural network based on the temperature coefficient, the positive sample similarity, and the negative sample similarity.

[0133] In one embodiment, the training module is further configured to: obtain model training requirement information corresponding to the target battery thermal runaway identification model; and construct the temperature coefficient corresponding to the Siamese neural network based on at least one of the model requirement accuracy and model training requirement duration represented by the model training requirement information.

[0134] In one embodiment, the training module is further configured to: extract short-time energy features from the training set in the positive samples to obtain short-time energy features; extract the number of times the training set in the positive samples crosses the zero level per unit time to obtain zero-crossing rate features; convert the training set in the positive samples from the time domain to the Mel spectrum domain to obtain spectral envelope features; and fuse the short-time energy features, zero-crossing rate features, and spectral envelope features to obtain training features.

[0135] In one embodiment, after updating the preliminary battery thermal runaway identification model using the validation set in the positive samples to obtain the target battery thermal runaway identification model, the training device for the battery thermal runaway identification model further includes: an application module, used to acquire the thermal runaway identification result output by the target battery thermal runaway identification model; outputting thermal runaway warning information when the thermal runaway occurrence probability represented by the thermal runaway identification result is greater than a first preset probability threshold and less than a second preset probability threshold; and cutting off the charging and discharging circuit of the battery corresponding to the thermal runaway identification result, controlling the fire protection system corresponding to the thermal runaway identification result to open, and outputting thermal runaway alarm information when the thermal runaway occurrence probability represented by the thermal runaway identification result is greater than the second preset probability threshold.

[0136] Each module in the training device for the aforementioned battery thermal runaway identification model can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0137] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a training method for a battery thermal runaway identification model. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0138] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0139] In one embodiment, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0140] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0141] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0142] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0143] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0144] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0145] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A training method for a battery thermal runaway identification model, characterized in that, The method includes: Acquire the sound signal during the thermal runaway process of the battery, and segment the sound signal to obtain various types of sound segmentation signals; Sound segments belonging to the type of safety valve opening sound are identified as positive samples in the training samples, and sound segments belonging to the type of noise are identified as negative samples in the training samples. A preliminary battery thermal runaway identification model is obtained by training using the training set from the positive samples and the negative samples. The preliminary battery thermal runaway identification model is updated using the validation set in the positive samples to obtain the target battery thermal runaway identification model.

2. The method according to claim 1, characterized in that, The preliminary battery thermal runaway identification model is obtained by training using the training set from the positive samples and the negative samples, including: Feature extraction is performed on the training set of the positive samples to obtain training features; Based on the training features and the negative samples, construct the loss function of the Siamese neural network; Based on the loss function, the twin neural network is trained to obtain a preliminary battery thermal runaway identification model.

3. The method according to claim 2, characterized in that, The step of constructing the loss function of the Siamese neural network based on the training features and the negative samples includes: The training features are extracted using a Siamese neural network to obtain positive sample features; and the negative samples are extracted using a Siamese neural network to obtain negative sample features. Obtain the temperature coefficient corresponding to the Siamese neural network, the positive sample similarity corresponding to the positive sample feature, and the negative sample similarity corresponding to the negative sample feature; The loss function of the Siamese neural network is constructed based on the temperature coefficient, the positive sample similarity, and the negative sample similarity.

4. The method according to claim 3, characterized in that, The step of obtaining the temperature coefficient corresponding to the Siamese neural network includes: Obtain the model training requirements information corresponding to the target battery thermal runaway identification model; The temperature coefficient corresponding to the Siamese neural network is constructed based on at least one of the model training requirement accuracy and model training requirement duration represented by the model training requirement information.

5. The method according to claim 2, characterized in that, The step of extracting features from the training set of the positive samples to obtain training features includes: Short-time energy features are extracted from the training set in the positive samples to obtain short-time energy features. The zero-crossing rate feature is obtained by extracting the number of times the training set in the positive samples crosses the zero level per unit time. The training set in the positive samples is converted from the time domain to the Mel spectrum domain to obtain the spectral envelope features; The training features are obtained by fusing the short-time energy features, the zero-crossing rate features, and the spectral envelope features.

6. The method according to any one of claims 1 to 5, characterized in that, After updating the preliminary battery thermal runaway identification model using the validation set from the positive samples to obtain the target battery thermal runaway identification model, the method further includes: Obtain the thermal runaway identification result output by the target battery thermal runaway identification model; If the probability of thermal runaway as represented by the thermal runaway identification result is greater than a first preset probability threshold and less than a second preset probability threshold, output thermal runaway warning information. If the probability of thermal runaway as represented by the thermal runaway identification result is greater than the second preset probability threshold, the charging and discharging circuit of the battery corresponding to the thermal runaway identification result is cut off, the fire protection system corresponding to the thermal runaway identification result is controlled to be turned on, and thermal runaway alarm information is output.

7. A training device for a battery thermal runaway identification model, characterized in that, The device includes: The acquisition module is used to acquire the sound signals during the thermal runaway process of the battery; The sample segmentation module is used to segment the sound signal to obtain various types of sound segmentation signals. The sound segmentation signals belonging to the safety valve opening sound type are determined as positive samples in the training samples, and the sound segmentation signals belonging to the noise type are determined as negative samples in the training samples. The training module is used to train a preliminary battery thermal runaway identification model using the training set in the positive samples and the negative samples; and to update the preliminary battery thermal runaway identification model using the validation set in the positive samples to obtain a target battery thermal runaway identification model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images