Battery fault type discrimination method and device, storage medium and electronic equipment

Abstract

The invention discloses a battery fault type judgment method and device, a storage medium and electronic equipment. The method comprises the steps that multi-mode target operation data of a target battery in a target period are acquired, and the multi-mode target operation data comprise gas data, deformation data, temperature data, sound data and electric power data; determining a target characteristic of the target battery in the target period based on the multi-modal target operation data; based on the target feature, risk probability distribution of the target battery in a predetermined time period is determined, the risk probability distribution comprises probabilities corresponding to multiple risk levels respectively, and the predetermined time period is a time period after the target period; and based on the risk probability distribution, determining a target early warning level and a target fault type of the target battery in the predetermined time period. According to the invention, the technical problem of inaccurate determination results of the early warning level and the fault type of the battery in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of battery technology, and more specifically, to a method, apparatus, storage medium, and electronic device for determining battery fault types. Background Technology

[0002] Lithium-ion batteries have been widely used in various fields due to their high energy density and long cycle life. However, safety hazards such as thermal runaway, lithium plating, and internal short circuits have always restricted their large-scale application. Therefore, efficient safety assessment and accurate fault type identification technologies have become core requirements. Currently, the safety assessment and accurate fault type identification technologies for lithium-ion batteries are mainly based on traditional monitoring technologies of the Battery Management System (BMS).

[0003] Traditional Battery Management System (BMS) technology focuses on single-parameter monitoring. It collects basic electrical and thermal parameters such as voltage, current, and surface temperature, and uses simple algorithms like threshold judgment and rule matching to achieve safety assessment and accurate fault type identification for lithium batteries. However, it has a weak ability to detect the complex physicochemical changes inside lithium batteries and cannot capture early signs of faults. Therefore, related technologies suffer from inaccurate determination of battery risk levels and fault types.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a method, apparatus, storage medium, and electronic device for determining battery fault types, in order to at least solve the technical problem of inaccurate determination results of battery warning levels and fault types in related technologies.

[0006] According to one aspect of the embodiments of this application, a method for determining the fault type of a battery is provided, comprising: acquiring multimodal target operating data of a target battery during a target period, wherein the multimodal target operating data includes gas data, deformation data, temperature data, sound data, and electrical data, wherein the gas data is generated by the electrolyte of the target battery during the chemical reaction, and the sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction; determining target characteristics of the target battery during the target period based on the multimodal target operating data, wherein the target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics, and electrical characteristics; determining the risk probability distribution of the target battery during a predetermined time period based on the target characteristics, wherein the risk probability distribution includes probabilities corresponding to multiple risk levels, and the predetermined time period is the time period after the target period; and determining the target warning level and target fault type of the target battery during the predetermined time period based on the risk probability distribution. This method achieves the goal of determining the target warning level and target fault type of the target battery by analyzing the target characteristics of the target battery obtained through feature extraction from the multimodal target operating data of the target battery, thereby improving the accuracy of the determination results of the target warning level and target fault type of the target battery.

[0007] Optionally, based on multimodal target operation data, the target features of the target battery in the target cycle are determined, including: using a target neural network model to extract features from the multimodal target operation data respectively, obtaining initial features corresponding to each of the multimodal target operation data, wherein the target neural network model includes multiple feature extraction branches, each corresponding one-to-one with the multimodal target operation data, and the convolution kernel size and learning rate of the feature extraction branches are determined according to the target operation data of the corresponding modality; concatenating the initial features corresponding to the multimodal target operation data respectively, obtaining the concatenated features of the target battery in the target cycle; and performing dimensionality reduction processing on the concatenated features to obtain the target features. By using different feature extraction branches to extract features from the target operation data of different modalities, the relevance and accuracy of the initial feature extraction results can be improved. At the same time, by performing dimensionality reduction processing on the concatenated features, the efficiency of the target warning level and target fault type determination process can be improved.

[0008] Optionally, the method further includes: determining the fluctuation intensity corresponding to each of the multimodal target operation data; and determining the convolution kernel size and learning rate corresponding to each of the multiple feature extraction branches based on the fluctuation intensity corresponding to each of the multimodal target operation data. By determining the convolution kernel size and learning rate of the feature extraction branches according to the target operation data of the corresponding modality, the targeting of the feature extraction process can be improved, thereby improving the quality of the initial feature extraction results and the accuracy of the target warning level and target fault type determination results.

[0009] Optionally, based on the risk probability distribution, the target warning level and target fault type of the target battery within a predetermined time period are determined. This includes: acquiring multimodal historical operating data of the test battery corresponding to various candidate fault types, wherein the test battery is of the same type as the target battery; based on the multimodal historical operating data, standard features corresponding to various candidate fault types are obtained by determining target features; the target warning level is determined based on the risk probability distribution; and the target fault type is determined based on the target warning level, target features, and standard features corresponding to various candidate fault types. By identifying the test battery of the same type as the target battery and the standard features corresponding to various candidate fault types, a reference basis is provided for determining the fault type, improving the accuracy and interpretability of the target fault type determination results.

[0010] Optionally, when the warning levels are divided into Level 1, Level 2, and Level 3, the target fault type is determined based on the target warning level, target features, and standard features corresponding to various candidate fault types. This includes: performing similarity analysis on the target features and the standard features corresponding to various candidate fault types to obtain multiple first similarity results, wherein each of the multiple first similarity results corresponds one-to-one with a variety of candidate fault types; when the target warning level indicates that the target battery is at Level 1 during a predetermined time period, the candidate fault type corresponding to the largest first similarity result among the multiple first similarity results is determined as the target fault type; or, when the target warning level indicates that the target battery is at Level 2 or Level 3 during a predetermined time period, the target fault type is determined based on the risk probability distribution and multiple first similarity results. In high-risk scenarios at the first-level warning level, the maximum matching decision is made directly based on the first similarity result between the target features and the standard features, ensuring decisive diagnosis and rapid response. In medium-risk scenarios at the second-level warning level or low-risk scenarios at the third-level warning level, the risk probability distribution and multiple first similarity results are combined to determine the target fault type, thereby improving the accuracy of the target fault type determination result.

[0011] Optionally, when multiple risk levels are involved, including high-risk, medium-risk, and low-risk levels, the target fault type is determined based on the risk probability distribution and multiple first similarity results. This includes: when the target warning level indicates that the target battery is at the second warning level within a predetermined time period, determining the first probability corresponding to the high-risk level and the second probability corresponding to the medium-risk level based on the risk probability distribution; obtaining multiple second similarity results based on the first probability, the second probability, and multiple first similarity results, wherein each of the multiple second similarity results corresponds one-to-one with multiple candidate fault types; and determining the candidate fault type corresponding to the largest second similarity result among the multiple second similarity results as the target fault type. Under the second warning level, by fusing the first probability and the second probability, the first similarity results are dynamically weighted and corrected to suppress misjudgments caused by sensor noise and operating condition fluctuations, thereby improving the accuracy of the target fault type determination result.

[0012] Optionally, when multiple risk levels are involved, including high-risk, medium-risk, and low-risk levels, the target fault type is determined based on the risk probability distribution and multiple first similarity results. This includes: when the target warning level indicates that the target battery is at the third warning level within a predetermined time period, determining the second probability corresponding to the medium-risk level and the third probability corresponding to the low-risk level based on the risk probability distribution; obtaining multiple third similarity results based on the second probability, the third probability, and multiple first similarity results, wherein each of the multiple third similarity results corresponds one-to-one with multiple candidate fault types; and determining the candidate fault type corresponding to the largest third similarity result among the multiple third similarity results as the target fault type. At the third warning level, the first similarity results are dynamically weighted and corrected by fusing the second and third probabilities, activating the early identification capability for latent faults such as lithium plating and slow SEI film decomposition, thereby improving the rationality and accuracy of the target fault type determination results.

[0013] According to another aspect of the embodiments of this application, a battery fault type discrimination device is provided, comprising: a data acquisition module, configured to acquire multimodal target operating data of a target battery during a target period, wherein the multimodal target operating data includes gas data, deformation data, temperature data, sound data, and electrical data, wherein the gas data is generated by the electrolyte of the target battery during the chemical reaction process, and the sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction process; a first determination module, configured to determine the target characteristics of the target battery during the target period based on the multimodal target operating data, wherein the target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics, and electrical characteristics; a second determination module, configured to determine the risk probability distribution of the target battery during a predetermined time period based on the target characteristics, wherein the risk probability distribution includes probabilities corresponding to multiple risk levels, and the predetermined time period is the time period after the target period; and a third determination module, configured to determine the target warning level and target fault type of the target battery during the predetermined time period based on the risk probability distribution.

[0014] According to another aspect of the embodiments of this application, a non-volatile storage medium is provided, which stores multiple instructions, the instructions being adapted for a battery fault type determination method to be loaded by a processor and any one of them executed.

[0015] According to another aspect of the embodiments of this application, an electronic device is provided, including: one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any one of the battery fault type determination methods.

[0016] According to another aspect of the embodiments of this application, a computer program product is provided, which, when executed on a data processing device, is adapted to perform steps of a battery fault type determination method.

[0017] In this embodiment, multimodal target operation data of the target battery during a target period is acquired. This multimodal target operation data includes gas data, deformation data, temperature data, sound data, and electrical data. Gas data is generated by the electrolyte of the target battery during the chemical reaction, and sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction. Based on this multimodal target operation data, target characteristics of the target battery during the target period are determined. These target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics, and electrical characteristics. Based on these target characteristics, a risk probability distribution of the target battery over a predetermined time period is determined. This risk probability distribution includes probabilities corresponding to various risk levels, and the predetermined time period is the time period following the target period. Based on the risk probability distribution, the target warning level and target fault type of the target battery during the predetermined time period are determined. This achieves the goal of determining the target warning level and target fault type of the target battery by analyzing the target characteristics obtained from feature extraction of the multimodal target operation data of the target battery. This improves the accuracy of the determination results for the target warning level and target fault type of the target battery, thereby solving the technical problem of inaccurate determination results for battery warning levels and fault types in related technologies. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0019] Figure 1 This is a flowchart of a battery fault type determination method provided according to an embodiment of this application;

[0020] Figure 2 This is a schematic diagram of an optional battery fault type identification system provided according to an embodiment of this application;

[0021] Figure 3 This is a structural diagram of an optional improved deep residual neural network evaluation model for batteries provided according to an embodiment of this application;

[0022] Figure 4 This is a flowchart of an optional battery fault type determination method provided according to an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of an optional battery fault type determination device provided according to an embodiment of this application. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] It should be noted that the information and data collected in this application (including but not limited to multimodal target operation data and multimodal historical operation data) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, and necessary confidentiality measures have been taken. These measures do not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding operation entry points for them to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0027] According to an embodiment of this application, a method embodiment for determining the fault type of a battery is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] Figure 1 This is a flowchart of a battery fault type determination method provided according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0029] Step S102: Obtain multimodal target operation data of the target battery during the target cycle. The multimodal target operation data includes gas data, deformation data, temperature data, sound data and power data. The gas data is generated by the electrolyte of the target battery during the chemical reaction process, and the sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction process.

[0030] It is understandable that by acquiring multimodal target operation data of the target battery during the target cycle, a rich and comprehensive data foundation can be provided for the subsequent determination of the target warning level and target fault type, thereby improving the rationality and accuracy of the determination results of the target warning level and target fault type.

[0031] Optionally, a sensor array can be used to acquire the aforementioned multimodal target operational data. The sensor array employs a distributed, three-dimensional deployment scheme, with targeted selection and layout based on the fault characteristics of the lithium battery (i.e., the target battery) to ensure accurate early signal acquisition. The multimodal sensor array may include gas sensor modules, strain gauge modules, thermocouple modules, acoustic microphone modules, and electrical parameter modules.

[0032] The gas sensor module employs a metal oxide hydrogen sensor, a thermally conductive dimethyl carbonate sensor, and an infrared carbon dioxide sensor, deployed on the top of the target battery pack and between individual cells to acquire gas data. The gas sensor module monitors characteristic gases such as H2 (hydrogen), DMC (dimethyl carbonate), and CO2 (carbon dioxide) generated by electrolyte evaporation, with a sampling frequency of 1 Hz and concentration detection accuracy down to the ppm level, to capture early gas generation signals.

[0033] The strain gauge module employs high-precision MEMS (Micro-Electro-Mechanical Systems) strain gauges, which are attached to the center of the side of the target battery cell to acquire deformation data. The strain gauge module is used to monitor minute deformations of the target battery during charging, discharging, and fault processes, achieving a deformation detection accuracy of 0.01 Pa and a sampling frequency of 10 Hz to capture deformation anomalies caused by lithium plating, internal structural damage, etc. Examples of these deformation data include micro-deformation of the casing caused by electrode expansion due to lithium plating, and thermal stress deformation caused by localized heat generation from internal short circuits.

[0034] The thermocouple module uses a combination of T-type and K-type thermocouples (i.e., temperature sensors) to acquire temperature data. The T-type thermocouple (copper-constantan) is deployed on the surface of the target battery, adaptable to a temperature range of -200~350℃, with an error of ≤±1℃, suitable for monitoring the temperature during normal operation; the K-type thermocouple (nickel-chromium-nickel-silicon) is deployed at key heat dissipation nodes of the target battery pack, withstanding high temperatures up to 1200℃, covering the monitoring of the thermal runaway stage, and has an acquisition frequency of 5Hz to achieve accurate capture of temperature distribution and rate of change.

[0035] The acoustic microphone module uses a MEMS microphone array (3-6 microphones) distributed inside the target battery pack to acquire sound data. The frequency response range is 20Hz-20kHz, and the acquisition frequency is 100Hz. It extracts characteristic frequency information by capturing abnormal acoustic signals such as electrolyte boiling, SEI film decomposition, and gas impact.

[0036] The electrical parameter module integrates high-precision voltage and current sensors and an internal resistance detection unit, which are precisely connected to the battery terminals to acquire power data. The module uses a 16-bit ADC (Analog-to-Digital Converter) chip for signal acquisition, with a sampling frequency of 1 kHz and a resolution ≥0.001V / 0.001A, to monitor voltage fluctuations, current anomalies, and internal resistance changes in real time.

[0037] All target operation data collected by the sensors are transmitted in dual mode via CAN (Controller Area Network) bus and Ethernet to achieve data synchronization control with a synchronization error of ≤10ms, ensuring time alignment of multimodal target operation data.

[0038] Step S104: Based on the multimodal target operation data, determine the target characteristics of the target battery in the target cycle, wherein the target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics and electrical characteristics;

[0039] It is understandable that feature extraction is performed on the acquired multimodal target operation data to determine the target characteristics of the target battery in the target cycle, including gas characteristics, deformation characteristics, temperature characteristics, sound characteristics, and electrical characteristics. By extracting features from the multimodal target operation data separately, a refined and differentiated perception of the target battery's operating state can be achieved, avoiding redundancy and noise interference caused by traditional "one-size-fits-all" feature extraction, and improving the accuracy and rationality of target features.

[0040] In one optional embodiment, determining the target features of the target battery in the target cycle based on multimodal target operation data includes: using a target neural network model to extract features from the multimodal target operation data to obtain initial features corresponding to each of the multimodal target operation data, wherein the target neural network model includes multiple feature extraction branches, each of which corresponds one-to-one with the multimodal target operation data, and the convolution kernel size and learning rate of the feature extraction branches are determined according to the target operation data of the corresponding modality; concatenating the initial features corresponding to the multimodal target operation data to obtain the concatenated features of the target battery in the target cycle; and performing dimensionality reduction processing on the concatenated features to obtain the target features.

[0041] It can be understood that multimodal target operation data is input into a target neural network model for feature extraction, resulting in initial features corresponding to each modality of the target operation data. The target neural network model includes multiple feature extraction branches, each used to extract features from the target operation data of its corresponding modality. The kernel size and learning rate of this feature extraction branch are determined based on the target operation data of that modality. The initial features corresponding to the multimodal target operation data are then concatenated to obtain the concatenated features of the target battery in the target cycle. To improve the efficiency of fault type discrimination, the concatenated features are dimensionality-reduced to obtain the target features of the target battery in the target cycle. By using different feature extraction branches to extract features from the target operation data of different modalities, the specificity and accuracy of the initial feature extraction results can be improved. Simultaneously, by performing dimensionality reduction on the concatenated features, the efficiency of the target warning level and target fault type determination process can be improved.

[0042] In an optional embodiment, the method further includes: determining the fluctuation intensity corresponding to each of the multimodal target running data; and determining the convolution kernel size and learning rate corresponding to each of the multiple feature extraction branches based on the fluctuation intensity corresponding to each of the multimodal target running data.

[0043] It is understandable that the kernel size and learning rate of the target feature extraction branch among multiple feature extraction branches are determined as follows: Based on the target modality's target running data corresponding to the target feature extraction branch, the fluctuation intensity of the target modality's target running data is determined, and based on this fluctuation intensity, the kernel size and learning rate of the target feature extraction branch are determined. By determining the kernel size and learning rate of the target feature extraction branch, the kernel size and learning rate corresponding to each of the multiple feature extraction branches are determined. By determining the kernel size and learning rate of the feature extraction branch based on the target running data of the corresponding modality, the targeting of the feature extraction process can be improved, thereby improving the quality of the initial feature extraction results and the accuracy of the target warning level and target fault type determination results.

[0044] Alternatively, the fluctuation intensity can be determined in the following way:

[0045]

[0046] Where W represents the number of data points included in the target running data of the i-th mode within the target period. The fluctuation intensity of the target running data for the i-th mode. For the target period, the k-th data point in the target running data of the i-th mode. Let be the standard deviation of the target running data for the i-th mode within the target period.

[0047] Optionally, a greater fluctuation intensity indicates more severe fluctuations in the target running data of the corresponding mode. Based on the fluctuation intensity and a pre-defined mapping relationship between fluctuation intensity and convolution kernel size, the convolution kernel size of the feature extraction branch used for feature extraction of the target running data of that mode is determined.

[0048] Optionally, the learning rate of the i-th feature extraction branch The following method is used to determine:

[0049]

[0050] in, As the baseline learning rate, To adjust the coefficient, ensure Satisfying 0.001≤ ≤0.02.

[0051] Optionally, a hierarchical fusion algorithm can be used to achieve preprocessing, feature extraction, and dimensionality reduction fusion of multimodal target running data. Specifically, it includes three stages: the first stage is GAN data preprocessing based on robust estimation; the second stage is adaptive fluctuation optimization neural network feature extraction; and the third stage is manifold mapping autoencoder neural network dimensionality reduction fusion.

[0052] Optionally, a Generative Adversarial Network (GAN) can be constructed, in which the generator learns the distribution of multimodal operating data of the target battery under normal operating conditions to generate high-quality synthetic data to supplement the training samples; the discriminator introduces a robust estimation function to identify and remove abnormal data in the multimodal target operating data collected by the multimodal sensor array due to sensor noise and environmental interference. At the same time, an interpolation algorithm is used to fill in the missing data in the multimodal target operating data, thereby standardizing and enhancing the multimodal target operating data and improving data quality.

[0053] Optionally, a neural network model based on an adaptive fluctuation optimization algorithm (i.e., a target neural network model) can be constructed. According to the characteristics of the target operating data of different modes (such as slow changes in gas concentration, instantaneous fluctuations in acoustic data, and other fluctuation intensity characteristics), the convolution kernel size and learning rate of the feature extraction branch of the neural network model can be dynamically adjusted. Time-domain features (rate of change, peak value), frequency-domain features (spectral peak value, bandwidth), and time-frequency-domain features (wavelet transform coefficients) are automatically extracted from the multi-modal target operating data to obtain the initial features corresponding to the multi-modal target operating data, so as to enhance the identification of early weak fault features of the target battery.

[0054] Optionally, a manifold mapping learning autoencoder can be constructed to map the extracted high-dimensional multimodal feature vectors (i.e., spliced ​​features represented in the form of vectors) to a low-dimensional manifold space, thereby mining the intrinsic correlation between different modal features. While retaining core fault information, the feature dimension can be reduced by more than 60%, thereby reducing the computational load of subsequent improvements to the deep residual neural network evaluation model and improving operational efficiency.

[0055] Step S106: Based on the target characteristics, determine the risk probability distribution of the target battery within a predetermined time period, wherein the risk probability distribution includes the probabilities corresponding to various risk levels, and the predetermined time period is the time period after the target period.

[0056] It is understandable that, based on the target characteristics, an improved deep residual neural network evaluation model is used to obtain the risk probability distribution of the target battery over a predetermined time period (e.g., the next seven days).

[0057] Optionally, the improved deep residual neural network evaluation model can adopt a network architecture of "input layer - residual block stacking layer - global average pooling layer - fully connected layer - output layer". The input layer is used to input the dimensionality-reduced and fused target feature vector (i.e., the target features represented in vector form). The residual block stacking layer contains 8-12 residual blocks (each residual block contains 2 convolutional layers, a batch normalization layer, and a ReLU (Rectified Linear Array) layer). The Unit (Modified Linear Unit) activation function solves the gradient vanishing problem through shortcut connections (skip connections), enhancing the model's deep feature learning ability; the global average pooling layer performs channel-wise average aggregation on the high-dimensional feature map output by the residual block stacking layer, compressing the features into a one-dimensional vector of the same length as the number of channels, thereby significantly reducing the parameter scale and improving the model's generalization ability, and suppressing overfitting; the fully connected layer maps the feature vector after global average pooling to the three-dimensional latent space, realizing the semantic transformation from deep features to risk levels; the output layer outputs the risk probability distribution of three risk levels—low risk, medium risk, and high risk—for the next 7 days (i.e., the predetermined time period).

[0058] Optionally, the training of the improved deep residual neural network evaluation model can be carried out using the full life cycle historical operating data of the test battery of the same type as the target battery (including data on normal operating conditions, lithium plating, internal short circuit, and thermal runaway precursors). The network parameters are initialized by transfer learning, and the Adam optimizer is used for iterative training for 100 rounds (learning rate 0.001, batch size 64). The model's generalization ability is improved by using the cross-entropy loss function and L2 regularization to prevent overfitting.

[0059] Step S108: Based on the risk probability distribution, determine the target warning level and target fault type of the target battery within a predetermined time period.

[0060] It is understood that, based on the risk probability distribution of the target battery within a predetermined time period, the target warning level and target fault type for the target battery within that predetermined time period are determined. The warning level may include, but is not limited to, Level 1, Level 2, and Level 3 warning levels, with Level 1 warning level being more severe than Level 2, and Level 2 warning level being more severe than Level 3. Fault types may include, but are not limited to, lithium plating, SEI (Solid Electrolyte Interphase) membrane decomposition, internal short circuit, thermal runaway, and electrolyte drying.

[0061] In one optional embodiment, determining the target warning level and target fault type of the target battery within a predetermined time period based on the risk probability distribution includes: acquiring multimodal historical operating data of the test battery corresponding to various candidate fault types, wherein the test battery is a battery of the same type as the target battery; obtaining standard features corresponding to various candidate fault types based on the multimodal historical operating data by using a method of determining target features; determining the target warning level based on the risk probability distribution; and determining the target fault type based on the target warning level, target features, and standard features corresponding to various candidate fault types.

[0062] The target warning level and target fault type of the target battery within a predetermined time period are determined using the following method based on the risk probability distribution of the target battery over that period. First, multimodal historical operating data for test batteries of the same type as the target battery are acquired under various candidate fault types. Then, by determining target features, standard features corresponding to each candidate fault type are obtained based on the multimodal historical operating data. Next, the target warning level for the target battery within the predetermined time period is determined based on the risk probability distribution of the target battery. Finally, based on the target warning level, combined with the target features and the standard features corresponding to each candidate fault type, the target fault type of the target battery within the predetermined time period is determined. By identifying the standard features corresponding to each candidate fault type for test batteries of the same type as the target battery, a reference basis is provided for fault type determination, improving the accuracy and interpretability of the target fault type determination results.

[0063] Optionally, the target warning level for the target battery within a predetermined time period can be determined using the following method, based on the risk probability distribution. If the predetermined time period is the next seven days, the obtained risk probability distribution includes the probability distribution results corresponding to each of the next seven days, with each probability distribution including the probabilities corresponding to multiple risk levels for that number of days. Based on the probability distribution results corresponding to the next seven days, combined with pre-set risk level determination rules (e.g., triggering a Level 1 warning when the first probability of a high-risk level is >80%, triggering a Level 2 warning when the first probability of a high-risk level is 20%-80%, and determining a low-risk level and triggering a Level 3 warning when the first probability of a high-risk level is <20%), the sub-warning levels corresponding to the next seven days are determined. If the sub-warning levels corresponding to the seven days include a Level 1 warning level, the number of days on which the Level 1 warning level first appears is determined as the target number of days. The target warning level for the target battery in the next seven days is determined to be Level 1. At the same time, based on the target probability distribution results of the target number of days (including the first probability of high risk level, the second probability of medium risk level, and the third probability of low risk level), as well as the target characteristics of the target number of days, and combined with the standard characteristics corresponding to various candidate fault types, the target fault type of the target battery in the next seven days is determined.

[0064] Optionally, if the first-level warning level does not appear in the sub-warning levels corresponding to the next seven days, the sub-warning levels corresponding to the next seven days are determined based on the probability distribution results for each of the next seven days, combined with pre-set risk level determination rules. If the sub-warning levels corresponding to the next seven days do not include the first-level warning level but include the second-level warning level, the number of days on which the second-level warning level first appears is determined as the target number of days. The target warning level for the target battery in the next seven days is determined to be the second-level warning level. At the same time, based on the target probability distribution results for the target number of days (including the first probability of high-risk level, the second probability of medium-risk level, and the third probability of low-risk level), and the target characteristics of the target number of days, combined with the standard characteristics corresponding to various candidate fault types, the target fault type for the target battery in the next seven days is determined.

[0065] Optionally, if neither Level 1 nor Level 2 warning levels appear in the sub-warning levels corresponding to the next seven days, the target warning level for the target battery in the next seven days is determined to be Level 3. The seven first probabilities corresponding to the high-risk level in the next seven days are compared, and the number of days corresponding to the maximum value is determined as the target number of days. Based on the risk probability distribution, the target risk probability result for the target number of days is determined (including the first probability for high-risk levels, the second probability for medium-risk levels, and the third probability for low-risk levels). Based on the above target risk probability result, the target characteristics of the target number of days, and combined with the standard characteristics corresponding to various candidate fault types, the target fault type for the target battery in the next seven days is determined.

[0066] Optionally, after receiving the target feature vector, the improved deep residual neural network evaluation model outputs the risk probability distribution of the target battery within a predetermined time period. When the first probability of a high-risk level is >80%, a first-level warning is triggered; when the first probability of a high-risk level is between 20% and 80%, a second-level warning is triggered; and when the first probability of a high-risk level is <20%, it is determined to be low-risk, triggering a third-level warning. Simultaneously, based on the aforementioned risk probability distribution, the target warning level of the target battery within the predetermined time period, as well as the risk evolution trend, are output, i.e., the changing trends of the sub-warning levels corresponding to multiple cycles within the predetermined time period. For example, if the predetermined time period is the next seven days, the obtained risk probability distribution includes the probability distribution results corresponding to each of the next seven days. Each probability distribution result includes the probabilities corresponding to multiple risk levels for the corresponding number of days. Based on the probability distribution results corresponding to each of the next seven days, combined with pre-set risk level determination rules, the sub-warning levels corresponding to each of the next seven days are determined, thereby obtaining the risk evolution trend of the target battery within the predetermined time period. Based on the aforementioned risk probability distribution and target warning level, the target fault type of the target battery within the predetermined time period (e.g., lithium plating, internal short circuit, etc.) is determined.

[0067] In one optional embodiment, when the warning level is divided into a first-level warning level, a second-level warning level, and a third-level warning level, the target fault type is determined based on the target warning level, target features, and standard features corresponding to various candidate fault types. This includes: performing similarity analysis on the target features and the standard features corresponding to various candidate fault types to obtain multiple first similarity results, wherein the multiple first similarity results correspond one-to-one with various candidate fault types; when the target warning level indicates that the target battery is at the first-level warning level during a predetermined time period, the candidate fault type corresponding to the largest first similarity result among the multiple first similarity results is determined as the target fault type; or, when the target warning level indicates that the target battery is at the second-level or third-level warning level during a predetermined time period, the target fault type is determined based on the risk probability distribution and the multiple first similarity results.

[0068] Understandably, when warning levels are divided into Level 1, Level 2, and Level 3, the target fault type is determined as follows: Similarity analysis is performed between the target features and the standard features corresponding to various candidate fault types, resulting in multiple first similarity results that correspond one-to-one with each candidate fault type. If the target warning level indicates that the target battery is at Level 1 within a predetermined time period, the target fault type is determined directly based on the multiple first similarity results; that is, the candidate fault type corresponding to the largest first similarity result among the multiple first similarity results is determined as the target fault type. If the target warning level indicates that the target battery is at Level 2 or Level 3 within a predetermined time period, the target fault type is determined based on the risk probability distribution and the multiple first similarity results. In high-risk scenarios at Level 1 warning level, the maximum matching decision is made directly based on the first similarity results between the target features and the standard features, ensuring decisive diagnosis and rapid response. In medium-risk scenarios at Level 2 warning level or low-risk scenarios at Level 3 warning level, the risk probability distribution and multiple first similarity results are combined to determine the target fault type, improving the accuracy of the target fault type determination.

[0069] Alternatively, the first similarity result can be determined in the following manner:

[0070]

[0071] in, This represents the first similarity result between the target feature and the m-th standard feature. For target features, The m-th standard feature corresponds to the m-th candidate fault type.

[0072] In one optional embodiment, when multiple risk levels, including high-risk, medium-risk, and low-risk levels, are considered, the target fault type is determined based on the risk probability distribution and multiple first similarity results. This includes: when the target warning level indicates that the target battery is at the second warning level within a predetermined time period, determining the first probability corresponding to the high-risk level and the second probability corresponding to the medium-risk level based on the risk probability distribution; obtaining multiple second similarity results based on the first probability, the second probability, and the multiple first similarity results, wherein each of the multiple second similarity results corresponds one-to-one with multiple candidate fault types; and determining the candidate fault type corresponding to the largest second similarity result among the multiple second similarity results as the target fault type.

[0073] It is understandable that if the target warning level indicates that the target battery is at the second warning level within a predetermined time period, then based on the risk probability distribution, a first probability corresponding to the high-risk level and a second probability corresponding to the medium-risk level are determined. Multiple first similarity results are then corrected based on the first and second probabilities to obtain multiple second similarity results. Finally, the candidate fault type corresponding to the largest second similarity result among the multiple second similarity results is determined as the target fault type. Under the second warning level, by fusing the first and second probabilities, the first similarity results are dynamically weighted and corrected to suppress misjudgments caused by sensor noise and operating condition fluctuations, thereby improving the accuracy of the target fault type determination result.

[0074] Alternatively, the second similarity result can be determined in the following manner:

[0075]

[0076] in, The second similarity result between the target feature and the m-th standard feature corresponds to the m-th candidate fault type. The first probability, This is the second probability.

[0077] In an optional embodiment, when multiple risk levels, including high-risk, medium-risk, and low-risk levels, are considered, the target fault type is determined based on the risk probability distribution and multiple first similarity results. This includes: when the target warning level indicates that the target battery is at the third warning level within a predetermined time period, determining the second probability corresponding to the medium-risk level and the third probability corresponding to the low-risk level based on the risk probability distribution; obtaining multiple third similarity results based on the second probability, the third probability, and the multiple first similarity results, wherein each of the multiple third similarity results corresponds one-to-one with multiple candidate fault types; and determining the candidate fault type corresponding to the largest third similarity result among the multiple third similarity results as the target fault type.

[0078] Understandably, if the target warning level indicates that the target battery is at the third warning level within a predetermined time period, then based on the risk probability distribution, the second probability corresponding to the medium-risk level and the third probability corresponding to the low-risk level are determined. Multiple first similarity results are then corrected based on the second and third probabilities to obtain multiple third similarity results. Finally, the candidate fault type corresponding to the largest third similarity result among the multiple third similarity results is determined as the target fault type. Under the third warning level, by fusing the second and third probabilities to dynamically weight and correct the first similarity results, the early identification capability for latent faults such as lithium plating and slow SEI film decomposition is activated, improving the rationality and accuracy of the target fault type determination results.

[0079] Alternatively, the third similarity result can be determined in the following manner:

[0080]

[0081] in, The third similarity result between the target feature and the m-th standard feature corresponds to the m-th candidate fault type. This is the third probability.

[0082] Optionally, if the predetermined time period includes a cycle, such as the next day, the target warning level and target fault type are determined based on the risk probability distribution of that cycle. If the predetermined time period includes multiple cycles, such as the target cycle being the current day and the predetermined time period being the next seven days, and the corresponding risk probability distribution includes the probability distribution results corresponding to the next seven days, the target probability distribution result corresponding to the target number of days in the determined next seven days (including the first probability of high risk level, the second probability of medium risk level, and the third probability of low risk level), as well as the target characteristics of the target number of days, are combined with the standard characteristics corresponding to the various candidate fault types to determine the target fault type.

[0083] Through the above steps S102 to S108, the target features of the target battery obtained by feature extraction based on the multimodal target operation data of the target battery can be analyzed to determine the target warning level and target fault type of the target battery. This achieves the technical effect of improving the accuracy of the determination results of the target warning level and target fault type of the target battery, thereby solving the technical problem of inaccurate determination results of the battery warning level and fault type in related technologies.

[0084] Based on the above embodiments and optional embodiments, this application proposes an optional implementation method for battery fault type discrimination. This optional implementation method can be understood as a lithium battery safety assessment method and system based on multimodal sensor fusion and artificial intelligence analysis.

[0085] To compensate for the limitations of single-parameter monitoring, existing technologies are beginning to explore multi-sensor integration. This involves introducing gas sensors, pressure sensors, strain sensors, and other sensors to construct a two-dimensional battery monitoring system based on either an "electric-thermal-gas" or "electric-thermal-mechanical" approach. Simple machine learning algorithms (such as LSTM (Long Short-Term Memory) network model and support vector machine model) are then used to optimize the early warning logic. Furthermore, by monitoring the concentrations of characteristic gases such as CO and H2, the early warning time can be improved by 5-15 minutes compared to traditional temperature monitoring, thus enhancing the timeliness of battery early warnings to some extent.

[0086] However, existing technologies still have the following shortcomings: Weak multi-source data fusion capabilities. They only achieve simple overlay or independent analysis of multi-sensor data, lacking a systematic fusion method for multimodal heterogeneous data ("gas-thermal-mechanical-acoustic-electrical"), failing to uncover the intrinsic correlations between different modalities, leading to incomplete feature extraction and misjudgments of target warning levels and target fault types; Insufficient warning timeliness. Existing technologies mostly rely on explicit parameters such as temperature and gas concentration, requiring the fault to develop to a mid-to-late stage before capturing obvious signals, with warning times generally limited to minutes, failing to provide sufficient time windows for fault handling and risk isolation, and making it difficult to prevent serious accidents such as thermal runaway; Poor adaptability of intelligent algorithms. Existing technologies mostly use general machine learning models, not optimized for the evolution of lithium battery faults, unable to effectively identify early, weak fault signs such as lithium plating and SEI film decomposition, and have weak adaptability to individual battery differences and operating condition fluctuations, with warning accuracy generally below 80%; Insufficient synergy between sensors and algorithms. In existing technologies, the deployment of sensors lacks targeted design, and the collected multimodal data cannot accurately match the algorithm requirements. At the same time, practical problems such as sensor noise and data loss are not considered, which affects the reliability of the target warning level and target fault type determination results and makes it difficult to adapt to complex application scenarios.

[0087] A lithium battery safety assessment method and system based on multimodal sensor fusion and artificial intelligence analysis, employing the aforementioned optional implementation methods, can solve the following technical problems: By constructing a multimodal sensor array and data fusion system adapted to lithium battery fault characteristics, efficient integration and in-depth mining of multimodal heterogeneous data are achieved, overcoming the limitations of single-parameter and simple superposition analysis; by optimizing the intelligent algorithm architecture, the ability to identify early weak fault symptoms of lithium batteries is improved, increasing the safety risk warning time from minutes to days, reserving sufficient time for risk handling; by enhancing the algorithm's adaptability to individual battery differences, operating condition fluctuations, and sensor noise, the accuracy of determining the warning target warning level and target fault type is increased to over 95%, reducing misjudgments; through the collaborative design of sensor deployment and algorithm models, the overall system reliability and scenario adaptability are improved, meeting the safety assessment needs of different lithium battery application scenarios.

[0088] Figure 2 This is a schematic diagram of an optional battery fault type discrimination system provided according to an embodiment of this application, such as... Figure 2As shown, the lithium battery safety assessment system based on multimodal sensor fusion and artificial intelligence analysis includes a multimodal sensor array module, a data transmission module, a multimodal data fusion module, a deep residual neural network evaluation module, an early warning output module, and a data storage module. These modules work together to determine the target early warning level and target fault type of the target battery throughout the entire process. Specifically, the sensor array module is responsible for multi-dimensional signal acquisition to obtain multimodal target operating data of the target battery during the target cycle; the data transmission module enables real-time synchronous transmission of multimodal target operating data; the multimodal data fusion module completes multimodal target operating data processing and feature fusion; the deep residual neural network evaluation module identifies and predicts risk probability distribution; the early warning output module outputs the target early warning level and target fault type; and the data storage module retains the multimodal target operating data and model parameters.

[0089] The multimodal sensor array module employs a distributed, three-dimensional deployment scheme, with sensor arrays selected and arranged specifically based on the fault characteristics of the lithium battery (i.e., the target battery) to ensure accurate early signal acquisition. The multimodal sensor array module includes a gas sensor module, a strain gauge module, a thermocouple module, an acoustic microphone module, and an electrical parameter module.

[0090] The gas sensor module employs a metal oxide hydrogen sensor, a thermally conductive dimethyl carbonate sensor, and an infrared carbon dioxide sensor, deployed on the top of the target battery pack and between individual cells to acquire gas data. The gas sensor module monitors characteristic gases such as H2 (hydrogen), DMC (dimethyl carbonate), and CO2 (carbon dioxide) generated by electrolyte evaporation, with a sampling frequency of 1 Hz and concentration detection accuracy down to the ppm level, to capture early gas generation signals.

[0091] The strain gauge module employs high-precision MEMS (Micro-Electro-Mechanical Systems) strain gauges, which are attached to the center of the side of the target battery cell to acquire deformation data. The strain gauge module is used to monitor minute deformations of the target battery during charging, discharging, and fault processes, achieving a deformation detection accuracy of 0.01 Pa and a sampling frequency of 10 Hz, to capture deformation anomalies caused by lithium plating, internal structural damage, and other reasons.

[0092] The thermocouple module uses a combination of T-type and K-type thermocouples (i.e., temperature sensors) to acquire temperature data. The T-type thermocouple (copper-constantan) is deployed on the surface of the target battery, adaptable to a temperature range of -200~350℃, with an error of ≤±1℃, suitable for monitoring the temperature during normal operation; the K-type thermocouple (nickel-chromium-nickel-silicon) is deployed at key heat dissipation nodes of the target battery pack, withstanding high temperatures up to 1200℃, covering the monitoring of the thermal runaway stage, and has an acquisition frequency of 5Hz to achieve accurate capture of temperature distribution and rate of change.

[0093] The acoustic microphone module uses a MEMS microphone array (3-6 microphones) distributed inside the target battery pack to acquire sound data. The frequency response range is 20Hz-20kHz, and the acquisition frequency is 100Hz. It extracts characteristic frequency information by capturing abnormal acoustic signals such as electrolyte boiling, SEI film decomposition, and gas impact.

[0094] The electrical parameter module integrates high-precision voltage and current sensors and an internal resistance detection unit, which are precisely connected to the battery terminals to acquire power data. The module uses a 16-bit ADC (Analog-to-Digital Converter) chip for signal acquisition, with a sampling frequency of 1 kHz and a resolution ≥0.001V / 0.001A, to monitor voltage fluctuations, current anomalies, and internal resistance changes in real time.

[0095] All target operation data collected by the sensors are transmitted in dual mode via the CAN (Controller Area Network) bus and Ethernet of the data transmission module to achieve data synchronization control with a synchronization error of ≤10ms, ensuring time alignment of multimodal target operation data.

[0096] The multimodal data fusion module employs a hierarchical fusion algorithm to achieve preprocessing, feature extraction, and dimensionality reduction fusion of multimodal target operational data. The specific process is as follows:

[0097] The first stage involves robust estimation-based GAN data preprocessing. A Generative Adversarial Network (GAN) is constructed, where the generator learns the distribution of multimodal operating data of the target battery under normal operating conditions to generate high-quality synthetic data to supplement the training samples. The discriminator introduces a robust estimation function to identify and remove abnormal data caused by sensor noise and environmental interference from the multimodal target operating data collected by the multimodal sensor array module. At the same time, an interpolation algorithm is used to fill in the missing data in the multimodal target operating data, thereby standardizing and enhancing the multimodal target operating data and improving data quality.

[0098] The second stage involves adaptive fluctuation optimization neural network feature extraction. A neural network model based on the adaptive fluctuation optimization algorithm (i.e., the target neural network model) is constructed. According to the characteristics of different modalities of the target operating data (such as slow changes in gas concentration, instantaneous fluctuations in acoustic data, and other fluctuation intensity characteristics), the kernel size and learning rate of the feature extraction branch of the neural network model are dynamically adjusted. Time-domain features (rate of change, peak value), frequency-domain features (spectral peak value, bandwidth), and time-frequency-domain features (wavelet transform coefficients) are automatically extracted from the multimodal target operating data to obtain the initial features corresponding to each multimodal target operating data, thereby enhancing the identification of early weak fault characteristics of the target battery.

[0099] The third stage involves dimensionality reduction and fusion using a manifold mapping autoencoder neural network. A manifold mapping learning autoencoder is constructed to map the extracted high-dimensional multimodal feature vectors (i.e., spliced ​​features represented in vector form) to a low-dimensional manifold space. This process uncovers the intrinsic relationships between different modal features, reducing the feature dimension by more than 60% while preserving core fault information. This reduces the computational load of subsequent improvements to the deep residual neural network evaluation model and improves operational efficiency.

[0100] The deep residual neural network evaluation module, based on the constructed improved deep residual neural network evaluation model, enables the identification and prediction of the risk probability distribution of the target battery.

[0101] Figure 3 This is a structural diagram of an optional improved deep residual neural network evaluation model for batteries provided according to an embodiment of this application, such as... Figure 3 As shown, the improved deep residual neural network evaluation model adopts a network architecture of "input layer - residual block stacking layer - global average pooling layer - fully connected layer - output layer". The input layer is used to input the dimensionality-reduced and fused target feature vector (i.e., the target features represented in vector form). The residual block stacking layer contains 8-12 residual blocks (each residual block contains 2 convolutional layers, a batch normalization layer, and a ReLU (Rectified Linear Array) layer). The Unit (Modified Linear Unit) activation function solves the gradient vanishing problem through shortcut connections (skip connections), enhancing the model's deep feature learning ability; the global average pooling layer performs channel-wise average aggregation on the high-dimensional feature map output by the residual block stacking layer, compressing the features into a one-dimensional vector of the same length as the number of channels, thereby significantly reducing the parameter scale and improving the model's generalization ability, and suppressing overfitting; the fully connected layer maps the feature vector after global average pooling to the three-dimensional latent space, realizing the semantic transformation from deep features to risk levels; the output layer outputs the risk probability distribution of three risk levels—low risk, medium risk, and high risk—for the next 7 days (i.e., the predetermined time period).

[0102] The improved deep residual neural network evaluation model is trained using historical operating data (including normal operating conditions, lithium plating, internal short circuit, and thermal runaway precursor scenarios) of the same test battery type as the target battery. The network parameters are initialized by transfer learning, and the Adam optimizer (learning rate 0.001, batch size 64) is used for 100 iterations of training. The cross-entropy loss function is combined with L2 regularization to prevent overfitting and improve the model's generalization ability.

[0103] The improved deep residual neural network evaluation model receives the target feature vector and outputs the risk probability distribution of the target battery over a predetermined time period. When the first probability of a high-risk level is >80%, a first-level warning is triggered; when the first probability of a high-risk level is between 20% and 80%, a second-level warning is triggered; and when the first probability of a high-risk level is <20%, it is determined to be low-risk, triggering a third-level warning. Simultaneously, based on the aforementioned risk probability distribution, the model outputs the target warning level of the target battery over the predetermined time period, as well as the risk evolution trend, i.e., the changing trends of the sub-warning levels corresponding to multiple periods within the predetermined time period. For example, if the predetermined time period is the next seven days, the obtained risk probability distribution includes the probability distribution results corresponding to each of the next seven days. Each probability distribution result includes the probabilities corresponding to multiple risk levels for the corresponding number of days. Based on the probability distribution results corresponding to each of the next seven days, combined with pre-set risk level determination rules, the sub-warning levels corresponding to each of the next seven days are determined, thus obtaining the risk evolution trend of the target battery over the predetermined time period. Based on the aforementioned risk probability distribution and target warning level, the target fault type of the target battery (e.g., lithium plating, internal short circuit, etc.) is determined within the predetermined time period.

[0104] Figure 4 This is a flowchart of an optional battery fault type determination method provided according to an embodiment of this application, such as... Figure 4 As shown, the steps of the lithium battery safety assessment method based on multimodal sensor fusion and artificial intelligence analysis include:

[0105] Step S1: Sensor array deployment and calibration. Deploy the multimodal sensors at preset locations on the target battery pack and individual cells, complete sensor calibration, ensure data acquisition accuracy meets standards, and establish the correspondence between sensors and individual battery cells;

[0106] Step S2: Real-time acquisition of multimodal target operation data. Gas concentration (i.e., gas data), deformation data, temperature data, acoustic signals (i.e., sound data), and electrical parameter data (i.e., power data) are synchronously collected by various sensors and transmitted to the data storage module and the multimodal data fusion module via the data transmission module.

[0107] Step S3: Multimodal data fusion processing. Process the multimodal target running data according to the process of "robust GAN preprocessing - adaptive feature extraction - manifold mapping dimensionality reduction" to obtain the fused feature vector (i.e., the target feature vector);

[0108] Step S4: Intelligent Safety Assessment. The fused feature vector is input into the improved deep residual neural network assessment model, which outputs the risk probability distribution for the next 7 days (i.e., the predetermined time period). Based on this risk probability distribution, the target warning level and target fault type of the target battery during the predetermined time period are determined.

[0109] Step S5: Tiered Early Warning Output. Based on the target early warning level, the result triggers the corresponding level of early warning, and the result is simultaneously fed back to the terminal device. At the same time, the historical database is updated for iterative optimization of the improved deep residual neural network evaluation model.

[0110] Step S6: Improve the online optimization of the deep residual neural network evaluation model. Regularly fine-tune the model parameters using newly added historical data to adapt to factors such as battery aging and changes in operating conditions, and maintain evaluation accuracy.

[0111] The above optional implementation methods achieve at least the following effects:

[0112] (1) By integrating multimodal target operation data from five dimensions and deep data mining, the limitations of traditional single-parameter monitoring can be overcome. Early weak fault signs such as lithium plating and SEI film decomposition can be captured, and the fault identification coverage reaches more than 90%, thus achieving a comprehensive assessment of the target battery status.

[0113] (2) Relying on the multimodal data fusion module and the deep residual neural network evaluation module, the early fault characteristics are accurately identified, and the warning time is increased from the traditional minute level to the day level. The warning of slow faults can be 3-7 days in advance, and the warning of fast faults can be several hours in advance, providing sufficient time for risk isolation and fault handling, and effectively reducing the incidence of safety accidents.

[0114] (3) Noise interference is eliminated by hierarchical data fusion algorithm. The optimized and improved deep residual neural network evaluation model is adapted to the target battery failure law. The early warning accuracy is stable at over 95% and the false alarm rate is less than 3%. At the same time, the collaborative design of sensors and algorithms improves the system's anti-interference ability and can be adapted to complex working conditions such as high and low temperatures and vibration.

[0115] (4) By using an online iterative optimization mechanism, the improved deep residual neural network evaluation model can adapt to battery aging, changes in operating conditions, etc., maintain high evaluation accuracy in the long term, reduce later maintenance costs, and provide technical support for the safety management of the target battery throughout its entire life cycle.

[0116] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0117] This embodiment also provides a battery fault type identification device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementations, or a combination of software and hardware, are also possible and contemplated.

[0118] According to an embodiment of this application, an apparatus embodiment for implementing a battery fault type determination method is also provided. Figure 5 This is a schematic diagram of a battery fault type determination device according to an embodiment of this application, such as... Figure 5 As shown, the above-mentioned battery fault type determination device includes a data acquisition module 502, a first determination module 504, a second determination module 506, and a third determination module 508. The device will be described below.

[0119] The data acquisition module 502 is used to acquire multimodal target operation data of the target battery during the target cycle. The multimodal target operation data includes gas data, deformation data, temperature data, sound data and power data. The gas data is generated by the electrolyte of the target battery during the chemical reaction process, and the sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction process.

[0120] The first determining module 504 is connected to the data acquisition module 502 and is used to determine the target characteristics of the target battery in the target cycle based on multimodal target operation data. The target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics and electrical characteristics.

[0121] The second determining module 506, connected to the first determining module 504, is used to determine the risk probability distribution of the target battery within a predetermined time period based on the target characteristics. The risk probability distribution includes the probabilities corresponding to various risk levels, and the predetermined time period is the time period after the target period.

[0122] The third determining module 508, connected to the second determining module 506, is used to determine the target warning level and target fault type of the target battery within a predetermined time period based on the risk probability distribution.

[0123] The battery fault type discrimination device provided in this application embodiment, by setting a data acquisition module 502, a first determination module 504, a second determination module 506, and a third determination module 508, achieves the purpose of analyzing the target features of the target battery obtained by feature extraction based on the multimodal target operation data of the target battery, determining the target warning level and target fault type of the target battery, thereby improving the accuracy of the determination results of the target warning level and target fault type of the target battery, and solving the technical problem of inaccurate determination results of battery warning level and fault type in related technologies.

[0124] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0125] It should be noted that the data acquisition module 502, the first determining module 504, the second determining module 506, and the third determining module 508 mentioned above correspond to steps S102 to S108 in the embodiments. The instances and application scenarios implemented by the above modules and their corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run in a computer terminal.

[0126] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0127] The aforementioned battery fault type determination device may further include a processor and a memory. The data acquisition module 502, the first determination module 504, the second determination module 506, the third determination module 508, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0128] The processor contains a core that retrieves the corresponding program unit from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0129] This application provides a non-volatile storage medium storing a program that, when executed by a processor, implements a method for determining the type of battery failure.

[0130] This application provides an electronic device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring multimodal target operating data of a target battery during a target cycle, wherein the multimodal target operating data includes gas data, deformation data, temperature data, sound data, and electrical data. The gas data is generated by the electrolyte of the target battery during the chemical reaction, and the sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction; based on the multimodal target operating data, determining the target characteristics of the target battery during the target cycle, wherein the target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics, and electrical characteristics; based on the target characteristics, determining the risk probability distribution of the target battery over a predetermined time period, wherein the risk probability distribution includes probabilities corresponding to various risk levels, and the predetermined time period is the time period following the target cycle; based on the risk probability distribution, determining the target warning level and target fault type of the target battery over the predetermined time period. The device in this document can be a server, PC, etc.

[0131] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program with the following method steps: acquiring multimodal target operating data of a target battery during a target period, wherein the multimodal target operating data includes gas data, deformation data, temperature data, sound data, and electrical data, wherein the gas data is generated by the electrolyte of the target battery during the chemical reaction, and the sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction; determining the target characteristics of the target battery during the target period based on the multimodal target operating data, wherein the target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics, and electrical characteristics; determining the risk probability distribution of the target battery during a predetermined time period based on the target characteristics, wherein the risk probability distribution includes probabilities corresponding to various risk levels, and the predetermined time period is the time period after the target period; and determining the target warning level and target fault type of the target battery during the predetermined time period based on the risk probability distribution.

[0132] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0134] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0135] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0136] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0137] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0138] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0139] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0140] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0141] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for determining the fault type of a battery, characterized in that, include: Acquire multimodal target operating data of the target battery during the target cycle, wherein the multimodal target operating data includes gas data, deformation data, temperature data, sound data and power data, wherein the gas data is generated by the electrolyte of the target battery during the chemical reaction, and the sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction; Based on the multimodal target operation data, the target characteristics of the target battery in the target cycle are determined, wherein the target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics, and electrical characteristics; Based on the target characteristics, the risk probability distribution of the target battery during a predetermined time period is determined, wherein the risk probability distribution includes probabilities corresponding to multiple risk levels, and the predetermined time period is the time period after the target period; Based on the risk probability distribution, the target warning level and target fault type of the target battery are determined during the predetermined time period.

2. The method according to claim 1, characterized in that, The step of determining the target characteristics of the target battery in the target cycle based on the multimodal target operation data includes: A target neural network model is used to extract features from the multimodal target running data to obtain initial features corresponding to the multimodal target running data. The target neural network model includes multiple feature extraction branches, each of which corresponds one-to-one with the multimodal target running data. The convolution kernel size and learning rate of the feature extraction branches are determined according to the target running data of the corresponding modality. The initial features corresponding to the multimodal target operation data are spliced ​​together to obtain the spliced ​​features of the target battery in the target cycle; The spliced ​​features are then subjected to dimensionality reduction processing to obtain the target features.

3. The method according to claim 2, characterized in that, The method further includes: Determine the fluctuation intensity corresponding to each of the multimodal target operation data; Based on the fluctuation intensity corresponding to the multimodal target running data, the convolution kernel size and learning rate corresponding to the multiple feature extraction branches are determined.

4. The method according to any one of claims 1 to 3, characterized in that, The step of determining the target warning level and target fault type of the target battery within the predetermined time period based on the risk probability distribution includes: Acquire multimodal historical operating data of the test battery under various candidate fault types, wherein the test battery is the same type as the target battery; Based on the multimodal historical operation data, the standard features corresponding to the various candidate fault types are obtained by determining the target features. Based on the aforementioned risk probability distribution, the target early warning level is determined; Based on the target warning level, the target characteristics, and the standard characteristics corresponding to the various candidate fault types, the target fault type is determined.

5. The method according to claim 4, characterized in that, When the warning levels are divided into Level 1, Level 2, and Level 3, determining the target fault type based on the target warning level, the target characteristics, and the standard characteristics corresponding to the various candidate fault types includes: The target features are compared with the standard features corresponding to the various candidate fault types to obtain multiple first similarity results, wherein the multiple first similarity results correspond one-to-one with the various candidate fault types; If the target warning level indicates that the target battery is at the first warning level during the predetermined time period, the candidate fault type corresponding to the largest first similarity result among the plurality of first similarity results is determined as the target fault type; or, When the target warning level indicates that the target battery is at the second or third warning level during the predetermined time period, the target fault type is determined based on the risk probability distribution and the plurality of first similarity results.

6. The method according to claim 5, characterized in that, When the multiple risk levels include high-risk, medium-risk, and low-risk levels, determining the target fault type based on the risk probability distribution and the multiple first similarity results includes: When the target warning level indicates that the target battery is at the second warning level during the predetermined time period, the first probability corresponding to the high risk level and the second probability corresponding to the medium risk level are determined based on the risk probability distribution. Based on the first probability, the second probability, and the plurality of first similarity results, a plurality of second similarity results are obtained, wherein the plurality of second similarity results correspond one-to-one with the plurality of candidate fault types; The candidate fault type corresponding to the largest second similarity result among the plurality of second similarity results is determined as the target fault type.

7. The method according to claim 5, characterized in that, When the multiple risk levels include high-risk, medium-risk, and low-risk levels, determining the target fault type based on the risk probability distribution and the multiple first similarity results includes: When the target warning level indicates that the target battery is at the third warning level during the predetermined time period, the second probability corresponding to the medium risk level and the third probability corresponding to the low risk level are determined based on the risk probability distribution. Based on the second probability, the third probability, and the plurality of first similarity results, a plurality of third similarity results are obtained, wherein the plurality of third similarity results correspond one-to-one with the plurality of candidate fault types; The candidate fault type corresponding to the largest third similarity result among the multiple third similarity results is determined as the target fault type.

8. A battery fault type identification device, characterized in that, include: The data acquisition module is used to acquire multimodal target operation data of the target battery during the target cycle. The multimodal target operation data includes gas data, deformation data, temperature data, sound data, and power data. The gas data is generated by the electrolyte of the target battery during the chemical reaction process, and the sound data is generated by the mechanical response triggered by the electrolyte during the chemical reaction process. The first determining module is used to determine the target characteristics of the target battery in the target cycle based on the multimodal target operation data, wherein the target characteristics include gas characteristics, deformation characteristics, temperature characteristics, sound characteristics and electrical characteristics; The second determining module is used to determine the risk probability distribution of the target battery over a predetermined time period based on the target characteristics, wherein the risk probability distribution includes probabilities corresponding to multiple risk levels, and the predetermined time period is the time period after the target period. The third determining module is used to determine the target warning level and target fault type of the target battery in the predetermined time period based on the risk probability distribution.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores multiple instructions, which are adapted to be loaded by a processor and executed by the battery fault type determination method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the battery fault type determination method according to any one of claims 1 to 7.

Images