A deep learning-based battery cell fault identification method and device

Abstract

The application discloses a kind of based on deep learning battery cell fault identification method and device, it is related to fault identification technical field.The method comprises: obtaining the battery cell in the running process of the to-be-identified multi-source monitoring data, and extracting the to-be-identified multi-source monitoring feature of the to-be-identified multi-source monitoring data;The to-be-identified multi-source monitoring feature is input into the battery cell fault identification model based on deep learning algorithm construction, carries out battery cell fault identification, obtains the fault probability vector of battery cell fault;According to the fault probability threshold of each fault category, the fault probability vector is disassembled, and the fault identification result of battery cell is obtained, and corresponding alarm signal is sent.The problems that the prior art exists that information dimension is single, fault signal sensitivity is insufficient and the identification ability for complex fault mode is limited are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault identification, in particular to a battery cell fault identification method and device based on deep learning. BACKGROUND

[0002] With the rapid development of electric vehicles, energy storage systems and other applications, the safety, reliability and life of power batteries have become a critical issue. As the core unit of the battery system, the operating state of the battery cell directly affects the performance and safety of the entire battery pack and even the entire system. During long-term operation, the battery cell may have various faults due to internal chemical reactions, external environmental influences or manufacturing defects, such as internal short circuit, external short circuit, thermal runaway, rapid capacity decay, etc. If these faults cannot be discovered and handled in time, they may lead to battery performance degradation, shortened life, and even serious safety accidents.

[0003] Traditional battery cell fault diagnosis methods mainly rely on voltage, current, temperature and other electrical signal data collected by the battery management system, and identify abnormalities through threshold setting or simple logical judgment. However, these methods often have limitations: first, the information dimension is single, and relying only on electrical signals may not fully reflect the internal state of the battery cell; second, the sensitivity to early and weak fault signals is insufficient, and false negatives are easy to occur; third, the recognition ability for complex fault patterns is limited. Therefore, developing a battery cell fault identification method that can integrate multi-source information and has high precision and reliability is of great practical significance to improve the safety and intelligence level of the battery system. SUMMARY

[0004] The present application provides a battery cell fault identification method and device based on deep learning, which solves the problems of single information dimension, insufficient fault signal sensitivity and limited recognition ability for complex fault patterns in the prior art.

[0005] In a first aspect, the present application provides a battery cell fault identification method based on deep learning, comprising:

[0006] Obtaining the battery cell's to-be-identified multi-source monitoring data during operation and extracting the to-be-identified multi-source monitoring features of the to-be-identified multi-source monitoring data;

[0007] Inputting the to-be-identified multi-source monitoring features into a battery cell fault identification model constructed based on a deep learning algorithm to perform battery cell fault identification and obtain a fault probability vector of the battery cell fault;

[0008] According to the fault probability threshold of each fault category, the fault probability vector is disassembled to obtain the battery cell's fault identification result and issue a corresponding alarm signal.

[0009] The technical scheme provided by the embodiments of the present application brings at least the following beneficial effects:

[0010] The multi-source information fusion technology combining audio signal data and electrical signal data is introduced, which can more comprehensively and accurately reflect the internal state of the battery cell and improve the robustness and accuracy of fault identification. The battery cell fault identification model constructed based on the deep learning algorithm can identify the deep relationship between the to-be-identified multi-source monitoring features and each fault category, providing the identification capability for complex fault patterns. Through the multi-source information fusion technology and the battery cell fault identification model, the internal slight abnormal changes of the battery cell can be perceived earlier, early fault warning is realized, valuable time is gained for taking maintenance or replacement measures, safety accidents are effectively prevented, the battery service life is prolonged, the threshold value is adaptively adjusted according to the actual operation state and risk level of the battery cell by using the fault probability threshold dynamic adjustment mechanism, the sensitivity and adaptability of identification are improved, and false positives and false negatives are reduced.

[0011] In an optional implementation, the to-be-identified multi-source monitoring data of the battery cell in the running process is acquired, and to-be-identified multi-source monitoring features of the to-be-identified multi-source monitoring data are extracted, including:

[0012] A microphone array is placed close to the battery cell, and audio signal data of the battery cell in the running process is collected;

[0013] A battery management system is used to collect electrical signal data of the battery cell in the running process, and the to-be-identified multi-source monitoring data are obtained in combination with the audio signal data; the electrical signal data include terminal voltage, charge and discharge current, and surface temperature;

[0014] The audio signal data in the to-be-identified multi-source monitoring data are preprocessed and feature-extracted to generate voiceprint features capable of reflecting the internal state changes of the battery cell;

[0015] The electrical signal data in the to-be-identified multi-source monitoring data are preprocessed and feature-extracted to extract electrical signal features reflecting the state of the battery cell, and the to-be-identified multi-source monitoring features are obtained in combination with the voiceprint features.

[0016] In an optional implementation, the audio signal data in the to-be-identified multi-source monitoring data are preprocessed and feature-extracted to generate voiceprint features capable of reflecting the internal state changes of the battery cell, including:

[0017] A noise suppression algorithm is used to perform noise reduction processing on the audio signal data in the to-be-identified multi-source monitoring data to obtain noise-reduced audio signal data;

[0018] The noise-reduced audio signal data are framed to obtain continuous frames of audio signals, and a Hanning window is applied to each frame of audio signal to obtain windowed continuous frames of audio signals.

[0019] The audio signal of each frame after windowing is subjected to a fast Fourier transform to obtain a frequency spectrum, and a voiceprint feature reflecting the internal state change of the battery cell is extracted according to the frequency spectrum.

[0020] In an optional implementation, the electrical signal data in the to-be-identified multi-source monitoring data is preprocessed and feature-extracted to extract an electrical signal feature reflecting the state of the battery cell, and the voiceprint feature is combined to obtain the to-be-identified multi-source monitoring feature, including:

[0021] The electrical signal data in the to-be-identified multi-source monitoring data is filtered and smoothed using a low-pass filter to obtain filtered electrical signal data;

[0022] The electrical signal feature reflecting the state of the battery cell of the filtered electrical signal data is extracted in a time window corresponding to the Hann window of the voiceprint feature;

[0023] The electrical signal feature and the voiceprint feature are integrated to obtain the to-be-identified multi-source monitoring feature.

[0024] In an optional implementation, the voiceprint feature includes a time-frequency domain feature, a time domain feature, and a statistical feature;

[0025] The time-frequency domain feature includes a spectral centroid, a spectral roll-off, and a mel-frequency cepstral coefficient;

[0026] The time domain feature includes a zero-crossing rate and an energy of the audio signal of each frame after windowing;

[0027] The statistical feature includes a time-frequency domain mean, a time-frequency domain standard deviation, a time-frequency domain maximum value, and a time-frequency domain minimum value obtained by statistical calculation of each time-frequency domain feature in the time window;

[0028] The filtered electrical signal data includes a filtered end voltage, a filtered charge-discharge current, and a filtered surface temperature after removal of high-frequency noise;

[0029] The electrical signal feature includes an average voltage, an average current, an average temperature, a voltage / current / temperature change rate, a voltage / current / temperature standard deviation, and a battery cell internal resistance.

[0030] In an optional implementation, the battery cell fault identification model is constructed based on a CNN-Attention-PC-LSTM-MLP algorithm, and the battery cell fault identification model includes an input layer, a feature fusion layer constructed based on a CNN-Attention algorithm, a physical constraint layer constructed based on a PC algorithm, a time series modeling layer constructed based on an LSTM algorithm, a battery cell fault identification layer constructed based on an MLP algorithm, and an output layer constructed based on a Softmax activation function, which are connected in sequence.

[0031] In an optional implementation, the sample data set composed of the normal state and several labeled multi-source monitoring data under different fault states is used for training to obtain the battery cell fault identification model, the training target is to minimize the total loss, and the model parameters of the battery cell fault identification model are updated;

[0032] The total loss function of the battery cell fault identification model includes a classification loss function and a physical constraint loss function;

[0033] The model parameters of the battery cell fault identification model are updated using the AdamW algorithm.

[0034] In an optional implementation, the multi-source monitoring features to be identified are input into the battery cell fault identification model constructed based on the deep learning algorithm to identify the battery cell fault, and a fault probability vector of the battery cell fault is obtained, including:

[0035] The input layer of the battery cell fault identification model constructed based on the deep learning algorithm receives the input multi-source monitoring features to be identified;

[0036] The feature fusion layer of the battery cell fault identification model is used to extract first high-level features of the electrical signal features and second high-level features of the voiceprint features in the multi-source monitoring features to be identified, and the first high-level features and the second high-level features are weighted and fused according to the dynamic attention weight to obtain the weighted fusion features;

[0037] The physical constraint layer of the battery cell fault identification model is used to correct the weighted fusion features output by the feature fusion layer to obtain the corrected weighted fusion features;

[0038] The time series modeling layer of the battery cell fault identification model is used to extract the time series features of the corrected weighted fusion features;

[0039] The battery cell fault identification layer of the battery cell fault identification model is used to identify the battery cell fault according to the time series features output by the time series modeling layer, and the Softmax activation function of the output layer is called to obtain the fault probability vector of the battery cell fault.

[0040] In an optional implementation, the fault probability vector is disassembled according to the fault probability threshold of each fault category to obtain the fault identification result of the battery cell, and a corresponding alarm signal is sent, including:

[0041] According to the deviation degree of the multi-source monitoring data to be identified and the several historical multi-source monitoring data, the anomaly entropy corresponding to the multi-source monitoring data to be identified is calculated;

[0042] According to the fault probability change of each fault category in the fault probability vector of the battery cell fault, the dynamic coefficient corresponding to each fault category is calculated;

[0043] According to the anomaly entropy of the to-be-identified multi-source monitoring data and the dynamic coefficient of each fault category, the fault probability threshold of each fault category is calculated;

[0044] According to the fault probability threshold of each fault category, the fault probability vector is disassembled, the fault category corresponding to the fault probability exceeding the fault probability threshold is summarized, and the fault recognition result of the battery cell is obtained, and a corresponding alarm signal is sent.

[0045] In a second aspect, an embodiment of the present application provides a battery cell fault recognition device based on deep learning, which is used to realize the battery cell fault recognition method. The device comprises:

[0046] A data acquisition unit is configured to acquire to-be-identified multi-source monitoring data of a battery cell during operation and extract to-be-identified multi-source monitoring features of the to-be-identified multi-source monitoring data;

[0047] A fault recognition unit is configured to input the to-be-identified multi-source monitoring features into a battery cell fault recognition model constructed based on a deep learning algorithm, perform battery cell fault recognition, and obtain a fault probability vector of a battery cell fault;

[0048] A fault disassembly unit is configured to disassemble the fault probability vector according to fault probability thresholds of each fault category, obtain a fault recognition result of the battery cell, and send a corresponding alarm signal.

[0049] In a third aspect, an embodiment of the present application provides an electronic device, which comprises:

[0050] at least one processor; and a memory connected with the at least one processor; wherein

[0051] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method provided in the first aspect of the present application.

[0052] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program. When the program is executed by a processor, the method provided in the first aspect of the present application is realized. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 FIG. 1 is a structural schematic diagram of an electronic device related to a hardware running environment of an embodiment of the present application;

[0054] Figure 2 FIG. 2 is a step flowchart of a battery cell fault recognition method based on deep learning provided by an embodiment of the present application;

[0055] Figure 3 FIG. 3 is a functional unit schematic diagram of a battery cell fault recognition device based on deep learning provided by an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0057] The scheme of the present application will be further described below with reference to the drawings.

[0058] Referring to Figure 1 , Figure 1 The electronic device structure schematic diagram of the hardware running environment involved in the embodiment scheme of the present application.

[0059] As Figure 1 shown, the electronic device can include a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 can include a display, an input unit such as a keyboard, and the optional user interface 1003 can further include a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory 1005 can be a high-speed random access memory (RAM) memory, or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a magnetic disk memory. The memory 1005 can also be an independent storage device from the aforementioned processor 1001.

[0060] Those skilled in the art can understand that Figure 1 the structure shown in the above does not constitute a limitation on the electronic device, and can include more or fewer components than the diagram, or combine certain components, or different component arrangements.

[0061] As Figure 1 shown, the memory 1005 as a storage medium can include an operating device, a data storage module, a network communication module, a user interface module, and an electronic program.

[0062] In Figure 1The network interface 1004 in the electronic device shown is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the electronic device of the present application can be arranged in the electronic device, and the electronic device calls the deep learning-based battery cell fault identification device stored in the memory 1005 through the processor 1001, and executes the deep learning-based battery cell fault identification method provided by the embodiments of the present application.

[0063] With reference to Figure 2 The embodiments of the present application provide a deep learning-based battery cell fault identification method, which can specifically include the following steps:

[0064] S201: Obtain the to-be-identified multi-source monitoring data of the battery cell in the running process, and extract the to-be-identified multi-source monitoring features of the to-be-identified multi-source monitoring data;

[0065] S202: Input the to-be-identified multi-source monitoring features into a battery cell fault identification model constructed based on a deep learning algorithm, perform battery cell fault identification, and obtain a fault probability vector of the battery cell fault;

[0066] S203: According to the fault probability threshold of each fault category, the fault probability vector is disassembled to obtain the fault identification result of the battery cell, and the corresponding alarm signal is sent.

[0067] The multi-source information fusion technology combining audio signal data and electrical signal data is introduced, which can more comprehensively and accurately reflect the internal state of the battery cell, and improve the robustness and accuracy of fault identification. The battery cell fault identification model constructed based on the deep learning algorithm can identify the deep relationship between the to-be-identified multi-source monitoring features and each fault category, and provide identification capability for complex fault modes. Through the multi-source information fusion technology and the battery cell fault identification model, the internal micro abnormal changes of the battery cell can be perceived earlier, early fault warning is realized, valuable time is gained for taking maintenance or replacement measures, thereby effectively preventing the occurrence of safety accidents, prolonging the service life of the battery, and adopting a fault probability threshold dynamic adjustment mechanism, so that the threshold can be adaptively adjusted according to the actual running state and risk level of the battery cell, improving the sensitivity and adaptability of identification, and reducing false positives and false negatives.

[0068] In an alternative implementation, obtaining the to-be-identified multi-source monitoring data of the battery cell in the running process, and extracting the to-be-identified multi-source monitoring features of the to-be-identified multi-source monitoring data, includes:

[0069] S2011: Use a microphone array placed near the battery cell to collect audio signal data of the battery cell in the running process; the sampling frequency is usually set to 44.1 kHz or higher to ensure the capture of high-frequency components. The original audio signal data collected is denoted as wherein is a time indicator;

[0070] S2012: Collecting the electrical signal data of the battery cell during operation using the battery management system, and obtaining the to-be-identified multi-source monitoring data in combination with the audio signal data; the electrical signal data includes terminal voltage, charging and discharging current, and surface temperature; the sampling frequency is usually 1 kHz or higher according to the configuration of the battery management system;

[0071] S2013: Preprocessing and feature extraction are performed on the audio signal data in the to-be-identified multi-source monitoring data to generate a voiceprint feature that can reflect the internal state change of the battery cell;

[0072] For the audio signal data, time-frequency domain, time domain and statistical features are extracted through noise reduction, frame division, windowing and Fourier transform, etc., so as to effectively capture the key information in the acoustic signal;

[0073] S2014: Preprocessing and feature extraction are performed on the electrical signal data in the to-be-identified multi-source monitoring data to extract an electrical signal feature reflecting the state of the battery cell, and to obtain the to-be-identified multi-source monitoring feature in combination with the voiceprint feature;

[0074] For the electrical signal data, statistical features and change rate features are extracted through filtering and smoothing, thereby enhancing the effectiveness of the electrical signal feature. These carefully extracted features lay a good foundation for the input of the subsequent deep learning model.

[0075] In an optional implementation, the preprocessing and feature extraction are performed on the audio signal data in the to-be-identified multi-source monitoring data to generate a voiceprint feature that can reflect the internal state change of the battery cell, including:

[0076] S20131: Using a noise suppression algorithm to perform noise reduction processing on the audio signal data in the to-be-identified multi-source monitoring data to obtain noise-reduced audio signal data ;

[0077] In this embodiment, the noise suppression algorithm includes adaptive filtering or wavelet transform method, and the adaptive filtering or wavelet transform method is used to remove the environmental noise and the non-fault-related noise inside the battery pack from the audio signal data.

[0078] For example, the improved Sym8 wavelet basis is used for multi-scale decomposition to remove the low-frequency and high-frequency noise components and retain the effective signal in the middle frequency band.

[0079] The formula is:

[0080]

[0081] In the formula, is a noise component determination threshold. is a signal standard deviation; is a sampling point number;

[0082] S20132: frame the noise-reduced audio signal data to obtain continuous frame audio signal, and apply a Hanning window to each frame of audio signal to obtain windowed continuous frame audio signal;

[0083] In this embodiment, the frame processing divides the noise-reduced audio signal data into short-time frames, each frame having a length of samples (for example, = 1024), a frame shift of samples (for example, M = 512), and a Hanning window applied to each frame to reduce spectral leakage; M The formula is:

[0084]

[0085]

[0086] In the formula, is the windowed continuous frame audio signal; is each frame of audio signal; is the Hanning window; n is a sample index; is a total number of samples;

[0087] S20133: perform a fast Fourier transform on each frame of windowed audio signal to obtain a frequency spectrum , and extract a voiceprint feature reflecting changes in the internal state of the battery cell according to the frequency spectrum.

[0088] In an alternative implementation, the electrical signal data in the to-be-identified multi-source monitoring data is preprocessed and feature extracted to extract an electrical signal feature reflecting the state of the battery cell, and the voiceprint feature is combined to obtain the to-be-identified multi-source monitoring feature, including:

[0089] S20141: filter and smooth the electrical signal data in the to-be-identified multi-source monitoring data using a low-pass filter to obtain filtered electrical signal data;

[0090] S20142: extract an electrical signal feature reflecting the state of the battery cell from the filtered electrical signal data within a time window corresponding to the Hanning window of the voiceprint feature;

[0091] S20143: integrate the electrical signal feature and the voiceprint feature to obtain the to-be-identified multi-source monitoring feature.

[0092] In this embodiment, the voiceprint feature includes a time-frequency domain feature, a time domain feature, and a statistical feature; ​

[0093] The time-frequency domain features include spectral centroid (reflecting the "brightness" of the sound), spectral rolloff (reflecting the spectral energy distribution), and mel-frequency cepstral coefficients (simulating the human auditory characteristics, capturing the timbre characteristics of the sound);

[0094] The formula is:

[0095]

[0096] In the formula, is the spectral centroid; is the frequency indicator; is the frame indicator;

[0097]

[0098] In the formula, is the spectral rolloff; is the frequency indicator; is the sum of all spectral energy whose frequency is less than or equal to ; is the total sum of spectral energy of all frequencies;

[0099]

[0100] In the formula, is the mel-frequency cepstral coefficient; is the energy output of the mel filter bank; is the discrete cosine transform function; the calculation steps of the mel-frequency cepstral coefficient include: mapping the spectrum to the mel filter bank, calculating the energy of each filter, taking the logarithm, and then performing the discrete cosine transform;

[0101] The time domain features include the zero-crossing rate and energy of the windowed audio signal of each frame;

[0102] The formula is:

[0103]

[0104] In the formula, is the zero-crossing rate; is the sign function, if >0, , <0, , if =0, ; is the windowed continuous frame of the audio signal of the sample;

[0105] ​​

[0106] wherein, is the energy; n is the sample indicator; is the total number of samples;

[0107] The statistical features include time-frequency domain mean, time-frequency domain standard deviation, time-frequency domain maximum value and time-frequency domain minimum value obtained by performing statistical calculation on each time-frequency domain feature within a time window;

[0108] The filtered electrical signal data includes filtered end voltage, filtered charge-discharge current and filtered surface temperature after removing high-frequency noise;

[0109] The electrical signal features include average voltage, average current, average temperature, voltage / current / temperature change rate, voltage / current / temperature standard deviation and cell internal resistance.

[0110] In an optional implementation, the cell fault identification model is constructed based on a Convolutional Neural Network (CNN)-Attention-Physical Constraint (PC)-Long Short-Term Memory (LSTM)-Multi-Layer Perceptron (MLP) algorithm, and the cell fault identification model includes an input layer, a feature fusion layer constructed based on a CNN-Attention algorithm, a physical constraint layer constructed based on a PC algorithm, a time series modeling layer constructed based on an LSTM algorithm, a cell fault identification layer constructed based on an MLP algorithm, and an output layer constructed based on a Softmax activation function, which are connected in sequence.

[0111] In this embodiment, the feature fusion layer includes two sub-networks constructed based on a CNN algorithm, which respectively extract higher-level features from the voiceprint features and the electrical signal features, and a dynamic attention mechanism is designed to dynamically adjust the fusion weight according to the importance of the current features.

[0112] The formula is:

[0113]

[0114] wherein, is the dynamic attention weight of the voiceprint features and the electrical signal features; is the activation function; is an MLP function, and the input is the spliced voiceprint features and the electrical signal features ; is the weighted fusion feature; a first high-level feature of the extracted electrical signal feature of the sub-network and a second high-level feature of the voiceprint feature; an indication quantity of a time window;

[0115] It is worth noting that the physical constraint layer is provided with a physical correlation model between the electrical signal running in the battery cell and the internal state change, which is used as prior knowledge constraint of the deep learning model to guide feature fusion and model training.

[0116] The formula of the physical correlation model is:

[0117]

[0118] In the formula, is a temperature change; is an average current, an internal resistance of the battery cell, and a voltage change rate; is a physical parameter related to the material and structure of the battery cell, which is usually calibrated through experiments;

[0119] The physical correlation model shows that the temperature rise is related to the square of the current, the internal resistance, and the voltage change rate. In the model training, the difference between the predicted temperature change of the model and the actually calculated temperature change can be used as part of the loss function.

[0120] In the embodiment, the sample data set composed of a plurality of labeled multi-source monitoring data under normal state and different fault states is used for training to obtain the battery cell fault recognition model. The training target is to minimize the total loss, and the model parameters of the battery cell fault recognition model are updated.

[0121] The total loss function of the battery cell fault recognition model includes a classification loss function and a physical constraint loss function.

[0122] The formula is:

[0123]

[0124] In the formula, is a total loss function; is a classification loss (such as cross-entropy loss); is a physical constraint loss function; is a hyperparameter for balancing the weight of the physical constraint term; is another regularization term;

[0125] The model parameters of the battery cell fault recognition model are updated using an Adam (Adam with Weight Decay, AdamW) optimizer with weight decay.

[0126] The formula of the AdamW optimizer is:

[0127]

[0128] wherein, is the adjusted operating condition perception learning rate; is the original operating condition perception learning rate; is the SOC change rate sensitivity coefficient (empirical value 0.5); is the battery state of charge (SOC) change rate;

[0129]

[0130] wherein, is the time t , t -1 neuron weight; is the first and second moment estimation of the gradient; is a very small constant used to prevent division by zero errors; is the weight decay coefficient, used to control the size of the weight to prevent overfitting;

[0131]

[0132] wherein, is the set of model parameters, which usually includes weight matrices, bias terms, etc., which determine the output and behavior of the model; is the training data set, which contains all the input data and corresponding labels used to train the model; is a sample extracted from the training data set; is the annotated multi-source monitoring data (features); is the corresponding fault state label (target value); is the total loss value, used to measure the difference between the model prediction and the true label; is the expected operation form, indicating that on the training data set all possible input-output pairs take the average, calculate the average value of the loss function on all samples;

[0133] In this embodiment, the multi-source monitoring features to be identified are input into the battery cell fault identification model constructed based on the deep learning algorithm to identify the battery cell fault, and a fault probability vector of the battery cell fault is obtained, including:

[0134] S2021: using the input layer of the battery cell fault identification model constructed based on the deep learning algorithm, receiving the input multi-source monitoring features to be identified;

[0135] S2022: using the feature fusion layer of the battery cell fault identification model, extracting the first high-level feature of the electrical signal feature and the second high-level feature of the voiceprint feature in the to-be-identified multi-source monitoring feature, and performing weighted fusion on the first high-level feature and the second high-level feature according to the dynamic attention weight to obtain a weighted fusion feature;

[0136] S2023: using the physical constraint layer of the battery cell fault identification model, correcting the weighted fusion feature output by the feature fusion layer to obtain a corrected weighted fusion feature;

[0137] S2024: using the time sequence modeling layer of the battery cell fault identification model, extracting the time sequence feature of the corrected weighted fusion feature;

[0138] S2025: using the battery cell fault identification layer of the battery cell fault identification model, performing battery cell fault identification according to the time sequence feature output by the time sequence modeling layer, and calling the Softmax activation function of the output layer to obtain a fault probability vector of the battery cell fault.

[0139] The formula is:

[0140]

[0141] In the formula, is the probability of belonging to the fault category ; is the fault category indicator; is the base; is the non-normalized score of the full connection layer output to ; is the non-normalized score of the full connection layer output to ; is the total number of fault categories;

[0142] In the present embodiment, the battery cell fault includes normal, internal short circuit tendency, thermal runaway tendency, active material attenuation, etc., and the fault probability vector of the battery cell fault is , wherein is the fault probability vector of the battery cell fault, is the probability of normal, internal short circuit tendency, thermal runaway tendency, and active material attenuation;

[0143] In an optional implementation, according to the fault probability threshold of each fault category, the fault probability vector is disassembled to obtain the fault identification result of the battery cell, and a corresponding alarm signal is sent, including:

[0144] S2031: calculating the anomaly entropy corresponding to the to-be-identified multi-source monitoring data according to the deviation degree of the to-be-identified multi-source monitoring data and a plurality of historical multi-source monitoring data;

[0145] The formula is:

[0146]

[0147] wherein, is the entropy of abnormality; is the first the distribution probability of the historical multi-source monitoring data; is the total number of data;

[0148] S2032: According to the change of the failure probability of each failure category in the failure probability vector of the battery cell failure, the dynamic coefficient corresponding to each failure category is calculated.

[0149] The formula is:

[0150]

[0151] wherein, is the dynamic coefficient of the failure category t at the moment ; is the failure probability of the moment t , t -1;

[0152] S2033: According to the entropy of abnormality of the multi-source monitoring data to be identified and the dynamic coefficient of each failure category, the failure probability threshold of each failure category is calculated.

[0153] The formula is:

[0154]

[0155] wherein, is the normalized failure probability threshold of the failure category ; is the original failure probability threshold; is the failure probability threshold reference value; is the entropy weight coefficient;

[0156] S2034: According to the failure probability threshold of each failure category, the failure probability vector is disassembled, the failure categories corresponding to the failure probability exceeding the failure probability threshold are summarized, the failure identification result of the battery cell is obtained, and the corresponding alarm signal (for example, lighting different color indicator lights, issuing sound alarm, sending alarm information to the monitoring system, etc.) is sent out.

[0157] For example, , and , the failure identification result of the battery cell includes internal short circuit tendency and active material attenuation.

[0158] The embodiment of the present application also provides a battery cell fault identification device based on deep learning. Figure 3 The embodiment of the present application also provides a battery cell fault identification device based on deep learning.

[0159] The data acquisition unit 301 is configured to acquire the to-be-identified multi-source monitoring data of the battery cell in the running process and extract to-be-identified multi-source monitoring features of the to-be-identified multi-source monitoring data.

[0160] The fault identification unit 302 is configured to input the to-be-identified multi-source monitoring features into a battery cell fault identification model constructed based on a deep learning algorithm, perform battery cell fault identification, and obtain a fault probability vector of the battery cell fault.

[0161] The fault disassembly unit 303 is configured to disassemble the fault probability vector according to fault probability thresholds of various fault categories, obtain a fault identification result of the battery cell, and send a corresponding alarm signal.

[0162] Based on the same inventive concept, another embodiment of the present application provides an electronic device, which comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus,

[0163] The memory is configured to store a computer program.

[0164] The processor is configured to execute the program stored on the memory, so as to realize the battery cell fault identification method based on deep learning.

[0165] The communication bus mentioned above can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The communication bus can be divided into an address bus, a data bus and a control bus. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus. The communication interface is configured to realize communication between the terminal and other devices. The memory can include a random access memory (RAM) and can also include a non-volatile memory, for example, at least one disk memory. Optionally, the memory can also be at least one storage device located away from the processor.

[0166] The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0167] In addition, to achieve the above object, an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the deep learning based battery cell fault identification method.

[0168] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, device or computer program product. Therefore, the embodiments of the present application can be in the form of a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present application can be in the form of a computer program product implemented on one or more computer usable vehicles (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer usable program codes.

[0169] The embodiments of the present application are described with reference to flowcharts and / or block diagrams according to the method, terminal device (apparatus) and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general purpose computer, a special purpose computer, an embedded processor or other programmable data processing terminal device to produce a machine, so that the instructions executed by the computer or other programmable data processing terminal device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one or more flows and / or blocks Figure 1 The device that implements the functions specified in one or more flows and / or blocks

[0170] These computer program instructions can also be stored in a computer readable storage medium, which can guide the computer or other programmable data processing terminal device to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices, which implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one or more flows and / or blocksFigure 1 the function specified in the one or more blocks.

[0171] These computer program instructions can also be loaded into computer or other programmable data processing terminal devices, so that a series of operation steps are performed on the computer or other programmable terminal devices to generate a computer-implemented process, so that the instructions executed on the computer or other programmable terminal devices provide a process for implementing the flowchart Figure 1 the flowchart or flowcharts and / or a block Figure 1 the steps of the function specified in the one or more blocks.

[0172] Finally, it should be noted that the relational terms herein, such as first and second, and the like, are used solely to distinguish one from another entity or action, without necessarily requiring or implying any actual relationship or order between such entities or actions. "And / or" means that either one of the two, or both, can be selected. Also, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, so that a process, method, article, or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or other elements inherent in such process, method, article, or terminal device. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or terminal device including the element.

[0173] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for identifying a cell failure based on deep learning, characterized in that, The method comprises the following steps: acquiring the to-be-identified multi-source monitoring data of the battery cell in the running process, and extracting the to-be-identified multi-source monitoring features of the to-be-identified multi-source monitoring data; inputting the to-be-identified multi-source monitoring features into a cell fault identification model constructed based on a deep learning algorithm to perform cell fault identification, and obtaining a fault probability vector of the cell fault; the cell fault identification model is constructed based on a CNN-Attention-PC-LSTM-MLP algorithm, and the cell fault identification model comprises an input layer, a feature fusion layer constructed based on a CNN-Attention algorithm, a physical constraint layer constructed based on a physical constraint PC algorithm, a time sequence modeling layer constructed based on an LSTM algorithm, a cell fault identification layer constructed based on an MLP algorithm, and an output layer constructed based on a Softmax activation function, which are connected in sequence; The method comprises the following steps: using the input layer of the cell fault identification model constructed based on the deep learning algorithm to receive the inputted to-be-identified multi-source monitoring features; using the feature fusion layer of the cell fault identification model to extract first high-level features of the electrical signal features and second high-level features of the voiceprint features in the to-be-identified multi-source monitoring features, and performing weighted fusion on the first high-level features and the second high-level features according to dynamic attention weights to obtain weighted fusion features; using the physical constraint layer of the cell fault identification model to correct the weighted fusion features output by the feature fusion layer to obtain corrected weighted fusion features; using the time sequence modeling layer of the cell fault identification model to extract time sequence features of the corrected weighted fusion features; using the cell fault identification layer of the cell fault identification model to perform cell fault identification according to the time sequence features output by the time sequence modeling layer, and calling the Softmax activation function of the output layer to obtain a fault probability vector of the cell fault; according to the fault probability threshold values of each fault category, the fault probability vector is disassembled to obtain the fault identification result of the cell, and the corresponding alarm signal is sent, comprising: calculating the anomaly entropy corresponding to the to-be-identified multi-source monitoring data according to the deviation degree of the to-be-identified multi-source monitoring data and a plurality of historical multi-source monitoring data; calculating the dynamic coefficient corresponding to each fault category according to the fault probability change of each fault category in the fault probability vector of the cell fault; calculating the fault probability threshold value of each fault category according to the anomaly entropy of the to-be-identified multi-source monitoring data and the dynamic coefficient of each fault category; according to the fault probability threshold values of each fault category, the fault probability vector is disassembled, the fault categories corresponding to the fault probability exceeding the fault probability threshold value are summarized, the fault identification result of the cell is obtained, and the corresponding alarm signal is sent. 2.The deep learning-based cell failure identification method of claim 1, wherein, acquiring the to-be-identified multi-source monitoring data of the battery cell in the running process, and extracting the to-be-identified multi-source monitoring features of the to-be-identified multi-source monitoring data, comprising: using a microphone array to be placed near the battery cell to collect audio signal data of the cell in the running process; The battery management system is used to collect electrical signal data of the battery cell during operation, and the audio signal data is combined to obtain the to-be-identified multi-source monitoring data; the electrical signal data includes terminal voltage, charging and discharging current, and surface temperature; The audio signal data in the to-be-identified multi-source monitoring data is preprocessed and feature extracted to generate a voiceprint feature capable of reflecting the internal state change of the battery cell; The electrical signal data in the to-be-identified multi-source monitoring data is preprocessed and feature extracted to extract an electrical signal feature reflecting the state of the battery cell, and the voiceprint feature is combined to obtain the to-be-identified multi-source monitoring feature. 3.The deep learning-based cell failure identification method of claim 2, wherein, The audio signal data in the to-be-identified multi-source monitoring data is preprocessed and feature extracted to generate a voiceprint feature capable of reflecting the internal state change of the battery cell, including: The noise suppression algorithm is used to perform noise reduction processing on the audio signal data in the to-be-identified multi-source monitoring data to obtain noise-reduced audio signal data; The noise-reduced audio signal data is framed to obtain continuous frame audio signal, and a Hanning window is applied to each frame of audio signal to obtain windowed continuous frame audio signal; Fast Fourier transform is performed on each frame of windowed audio signal to obtain a frequency spectrum, and a voiceprint feature capable of reflecting the internal state change of the battery cell is extracted according to the frequency spectrum.

4. The deep learning-based cell failure identification method of claim 3, wherein, The electrical signal data in the to-be-identified multi-source monitoring data is preprocessed and feature extracted to extract an electrical signal feature reflecting the state of the battery cell, and the voiceprint feature is combined to obtain the to-be-identified multi-source monitoring feature, including: The low-pass filter is used to filter and smooth the electrical signal data in the to-be-identified multi-source monitoring data to obtain filtered electrical signal data; In a time window corresponding to the Hanning window of the voiceprint feature, an electrical signal feature reflecting the state of the battery cell is extracted from the filtered electrical signal data; The electrical signal feature and the voiceprint feature are integrated to obtain the to-be-identified multi-source monitoring feature.

5. The deep learning-based cell failure identification method of claim 4, wherein, The voiceprint feature includes time-frequency domain feature, time domain feature, and statistical feature; The time-frequency domain feature includes spectral centroid, spectral roll-off, and mel-frequency cepstral coefficient; The time domain feature includes the zero-crossing rate and energy of each frame of windowed audio signal; The statistical feature includes time-frequency domain mean, time-frequency domain standard deviation, time-frequency domain maximum value, and time-frequency domain minimum value obtained by statistical calculation of each time-frequency domain feature in a time window; The filtered electrical signal data includes filtered terminal voltage, filtered charging and discharging current, and filtered surface temperature after removing high-frequency noise; The electrical signal feature includes average voltage, average current, average temperature, voltage / current / temperature change rate, voltage / current / temperature standard deviation, and battery internal resistance.

6. The deep learning-based cell failure identification method of claim 1, wherein, A sample data set composed of a plurality of labeled multi-source monitoring data under normal state and different fault states is used for training to obtain a battery cell fault recognition model, the training target is to minimize the total loss, and the model parameters of the battery cell fault recognition model are updated; The total loss function of the battery cell fault recognition model includes a classification loss function and a physical constraint loss function; updating the model parameters of the cell failure identification model using an AdamW algorithm.

7. A deep learning-based cell fault identification device for implementing the cell fault identification method according to any one of claims 1-6, characterized in that, The device comprises: a data acquisition unit configured to acquire a plurality of to-be-identified monitoring data of a battery cell during operation and extract a plurality of to-be-identified monitoring features of the plurality of to-be-identified monitoring data; a failure identification unit configured to input the plurality of to-be-identified monitoring features into a cell failure identification model constructed based on a deep learning algorithm, perform cell failure identification, and obtain a failure probability vector of the cell failure; a failure disassembly unit configured to disassemble the failure probability vector according to a failure probability threshold of each failure category, obtain a failure identification result of the cell, and send a corresponding alarm signal.

Images