Intelligent battery management method and device, equipment and storage medium

Abstract

The invention relates to the technical field of battery management, and discloses an intelligent battery management method and device, equipment and a storage medium. The method comprises the following steps: performing multi-source data acquisition and feature separation on a first sensing parameter and a second sensing parameter of a target intelligent battery to obtain a basic state feature, and performing frequency domain decomposition and space-time reconstruction on the basic state feature to obtain battery multi-modal data; deep coupling of electrochemical features and structural features is achieved through multi-level combination feature extraction and fusion, a fusion feature vector is obtained, separation of two sensing parameter channels, calculation of dual-channel weight and prediction of a multi-target time sequence are carried out on the fusion feature vector, and a multi-dimensional state evaluation index is obtained; and obtaining a comprehensive state evaluation index of the battery, thereby topping the corresponding fault risk level based on the comprehensive state evaluation index, and finally generating a battery management result of the target intelligent battery. And the accuracy and reliability of state evaluation and fault diagnosis of the intelligent battery are improved.

Description

Technical Field

[0001] The present invention relates to the field of battery management technology, and in particular to an intelligent battery management method, device, equipment and storage medium. Background Art

[0002] In the field of smart battery technology, battery safety and reliability assessment is an important part of R&D and quality control, and real-time monitoring and fault diagnosis of battery status is the most challenging stage. Accurately assessing the operating status of battery packs is crucial for manufacturers, R&D institutions and testing centers, as it directly affects battery performance optimization, life prediction and safety management.

[0003] At present, the commonly used battery management methods are mainly based on the monitoring and analysis of traditional electrochemical parameters, such as collecting data such as voltage, current, temperature, etc. to establish fault models, or using machine learning technology to evaluate the battery status. However, these methods still face challenges in dealing with multi-dimensional parameter fusion, nonlinear feature extraction, and dynamic response analysis. In particular, the internal structural evolution characteristics of the battery, such as mechanical stress changes, material interface evolution, structural integrity and other non-traditional parameters, are neglected. These factors have an important impact on battery status assessment and fault diagnosis. That is, the existing battery management methods fail to fully utilize the collaborative analysis of multi-source sensor data, resulting in low accuracy and reliability of status assessment and fault diagnosis. Summary of the invention

[0004] The main purpose of the present invention is to solve the problem that the existing battery management method fails to fully utilize the collaborative analysis of multi-source sensor data, resulting in low accuracy and reliability of state assessment and fault diagnosis.

[0005] A first aspect of the present invention provides an intelligent battery management method, the intelligent battery management method comprising: collecting first multi-source original sensor parameter data corresponding to a first sensor parameter and second multi-source original sensor parameter data corresponding to a second sensor parameter of a target intelligent battery, and performing spatiotemporal correlation and multiple preprocessing on the first multi-source original sensor parameter data and the second multi-source original sensor parameter data, respectively, to obtain first multimodal sensor data and second multimodal sensor data; extracting multi-level combined features from the first multimodal sensor data and the second multimodal sensor data to obtain a multi-battery parameter combination feature set, and performing multi-dimensional feature fusion on each of the battery parameter combination feature sets to obtain a fused feature vector of the intelligent battery state; performing a first The multi-source sensing parameter channel and the second multi-source sensing parameter channel are separated to obtain an electrochemical state characteristic flow and a structural evolution characteristic flow, and the dual-channel cross-attention weights are calculated and the multi-target timing is predicted for the electrochemical state characteristic flow and the structural evolution characteristic flow to obtain a multi-dimensional state evaluation index of the target smart battery; the timing anomaly detection and abnormal spatial distribution are determined for the multi-dimensional state evaluation index to obtain a multi-dimensional fault feature vector, and the propagation path calculation and risk level division are performed on the multi-dimensional fault feature vector to obtain the fault risk level of the target smart battery; the fault risk level and the multi-dimensional state evaluation index are calculated by safety threshold grading and converted by management instructions to generate a battery management result of the target smart battery.

[0006] Optionally, in a first implementation of the first aspect of the present invention, the first multi-source original sensor parameter data and the second multi-source original sensor parameter data are respectively subjected to spatiotemporal correlation and multiple preprocessing to obtain first multimodal sensor data and second multimodal sensor data, including: frequency segmenting the first multi-source original sensor parameter data to obtain electrochemical low-frequency data segments, electrochemical medium-frequency data segments and electrochemical high-frequency data segments, and frequency segmenting the second multi-source original sensor parameter data to obtain mechanical low-frequency data segments, mechanical medium-frequency data segments and mechanical high-frequency data segments; frequency segmenting the electrochemical low-frequency data segments and the mechanical low-frequency data segments. The electrochemical intermediate frequency data segment is subjected to Kalman filtering to obtain a basic state feature, and the electrochemical intermediate frequency data segment and the mechanical intermediate frequency data segment are subjected to wavelet multi-scale decomposition to obtain a dynamic response feature, and the electrochemical high frequency data segment and the mechanical high frequency data segment are subjected to empirical mode decomposition to obtain a transient evolution feature; the basic state feature, the dynamic response feature and the transient evolution feature are synchronously matched in time and space to obtain initial multimodal sensing data, and the initial multimodal sensing data is subjected to signal-to-noise ratio evaluation and weighted calculation of the corresponding original sensing parameters to obtain first multimodal sensing data and second multimodal sensing data.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the multi-level combined feature extraction of the first multimodal sensing data and the second multimodal sensing data to obtain a multi-battery parameter combination feature set, and multi-dimensional feature fusion of each of the battery parameter combination feature sets to obtain a fusion feature vector of the intelligent battery state includes: performing hierarchical feature extraction on the first multimodal sensing data to obtain a voltage and current feature set and a temperature internal resistance feature set, and performing hierarchical feature extraction on the second multimodal sensing data to obtain a vibration acoustic emission feature set and an ultrasonic displacement feature set; performing time-series pairing on the voltage and current feature set and the vibration acoustic emission feature set to obtain a multi-dimensional feature relationship matrix, and performing cross-correlation calculation and nonlinear mapping on the multi-dimensional feature relationship matrix to obtain a first battery parameter combination feature set; performing spatial registration on the temperature internal resistance feature set and the ultrasonic displacement feature set. A structural parameter relationship matrix is ​​obtained, and time-frequency feature mapping is performed on the structural parameter relationship matrix to obtain a second battery parameter combination feature set; wavelet packet multi-level decomposition is performed on the first battery parameter combination feature set to obtain a multi-scale battery reaction stress feature, and empirical mode decomposition is performed on the second battery parameter combination feature set to obtain a multi-scale thermal field deformation feature, and energy reconstruction and feature selection are performed on the multi-scale battery reaction stress feature to obtain an electrochemical strain coupling feature, and modal screening and feature fusion are performed on the multi-scale thermal field deformation feature to obtain a temperature deformation coupling feature; hierarchical contribution calculation is performed on the electrochemical strain coupling feature and the temperature deformation coupling feature to obtain a multiple coupling weight matrix, and based on the multiple coupling weight matrix, the electrochemical strain coupling feature and the temperature deformation coupling feature are weightedly fused to obtain a fused feature vector of the intelligent battery state.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the separation of the first multi-source sensing parameter channel and the second multi-source sensing parameter channel of the fused feature vector to obtain an electrochemical state feature stream and a structural evolution feature stream includes: calculating the attention weight of the fused feature vector corresponding to the first multi-source sensing parameter, extracting the initial electrochemical feature, and compressing the feature dimension and reconstructing the electrochemical feature of the initial electrochemical feature to obtain an electrochemical feature subset of the target smart battery; extracting the dual-branch temporal convolution feature of the electrochemical feature subset to obtain an electrochemical response. Matrix, and reconstruct the electrochemical response matrix in time series and enhance the performance characteristics to obtain the electrochemical state characteristic flow; calculate the local similarity of the fused feature vector corresponding to the second multi-source sensing parameter, extract the initial structural features of the corresponding characteristic components, reconstruct the structural features and screen the key structural features of the initial structural features to obtain the structural evolution feature subset of the target smart battery; extract the dual-branch spatial convolution features of the structural evolution feature subset to obtain the structural response matrix, and perform spatial reconstruction and feature enhancement on the structural response matrix to obtain the structural evolution characteristic flow.

[0009] Optionally, in a fourth implementation manner of the first aspect of the present invention, the electrochemical state characteristic flow and the structural evolution characteristic flow are subjected to dual-channel cross-attention weight calculation and multi-target timing prediction to obtain a multi-dimensional state evaluation index of the target smart battery, including: performing self-attention weight calculation in the timing dimension on the corresponding charge and discharge characteristic branches and polarization characteristic branches in the electrochemical state characteristic flow to obtain electrochemical enhancement characteristics, and performing self-attention weight calculation in the spatial dimension on the corresponding stress characteristic branches and deformation characteristic branches in the structural evolution characteristic flow to obtain structural enhancement characteristics; performing dual-channel cross-attention mapping calculation on the electrochemical enhancement characteristics and the structural enhancement characteristics to obtain a feature weight mapping matrix, and performing sliding timing segmentation and battery performance trend prediction on the feature weight mapping matrix to obtain a state evolution sequence; based on a preset branch evaluation model, performing multi-branch state evaluation on the state evolution sequence to obtain an initial battery state evaluation value, and performing confidence calibration and index optimization on the initial battery state evaluation value to obtain a multi-dimensional state evaluation index of the target smart battery.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, the multidimensional state evaluation index is subjected to time series anomaly detection and determination of abnormal spatial distribution to obtain a multidimensional fault feature vector, and the multidimensional fault feature vector is subjected to propagation path calculation and risk level division to obtain the fault risk level of the target smart battery, including: performing deviation calculation of preset multiple monitoring parameters on the multidimensional state evaluation index to obtain a state change sequence, and performing multi-level battery abnormal working threshold classification and abnormal location on the state change sequence to obtain an abnormal state sequence; performing spatial distribution mapping and associated feature extraction on the abnormal state sequence to obtain a multidimensional fault feature vector, and performing fault association network construction on the multidimensional fault feature vector to obtain an initial propagation path matrix; performing diffusion rate and propagation direction calculation on the initial propagation path matrix to obtain fault dynamic evolution parameters, and performing spatiotemporal sequence reconstruction on the fault dynamic evolution parameters to obtain a battery fault propagation matrix; performing multidimensional risk indicator calculation on the fault propagation matrix to obtain a fault risk score, and performing level division and priority sorting on the fault risk score to obtain the fault risk level of the target smart battery.

[0011] Optionally, in a sixth implementation method of the first aspect of the present invention, the fault risk level and the multi-dimensional state evaluation index are subjected to safety threshold grading calculation and management instruction conversion to generate a battery management result of the target smart battery, including: safety factor mapping and working range conversion of the fault risk level and the multi-dimensional state evaluation index to obtain battery working constraint parameters; charging and discharging parameter conversion of the battery working constraint parameters corresponding to the power limit coefficient to obtain power management instructions, and cooling channel flow distribution and heat dissipation power distribution calculation corresponding to the temperature threshold of the battery working constraint parameters to obtain temperature management instructions, and active balancing circuit parameter calculation corresponding to the single cell state difference of the battery working constraint parameters to obtain balancing management instructions; priority sorting and battery state regulation of the power management instructions, the temperature management instructions and the balancing management instructions to generate the battery management result of the target smart battery.

[0012] The second aspect of the present invention provides an intelligent battery management device, which includes: a preprocessing module, which is used to collect first multi-source original sensor parameter data corresponding to a first sensor parameter of a target intelligent battery and second multi-source original sensor parameter data corresponding to a second sensor parameter, and respectively perform spatiotemporal correlation and multiple preprocessing on the first multi-source original sensor parameter data and the second multi-source original sensor parameter data to obtain first multimodal sensor data and second multimodal sensor data; a feature fusion module, which is used to extract multi-level combined features of the first multimodal sensor data and the second multimodal sensor data to obtain a multi-battery parameter combination feature set, and perform multi-dimensional feature fusion on each of the battery parameter combination feature sets to obtain a fused feature vector of the intelligent battery state; an indicator evaluation module, which is used to evaluate the fused feature vector The first multi-source sensing parameter channel and the second multi-source sensing parameter channel are separated to obtain an electrochemical state characteristic flow and a structural evolution characteristic flow, and the dual-channel cross-attention weights are calculated and the multi-target timing is predicted for the electrochemical state characteristic flow and the structural evolution characteristic flow to obtain a multi-dimensional state evaluation index of the target smart battery; a risk classification module is used to perform timing anomaly detection and abnormal spatial distribution determination on the multi-dimensional state evaluation index to obtain a multi-dimensional fault feature vector, and propagation path calculation and risk level classification are performed on the multi-dimensional fault feature vector to obtain a fault risk level of the target smart battery; an instruction conversion module is used to perform safety threshold grading calculation and management instruction conversion on the fault risk level and the multi-dimensional state evaluation index to generate a battery management result of the target smart battery.

[0013] The third aspect of the present invention provides an intelligent battery management device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the intelligent battery management device executes the various steps of the above-mentioned intelligent battery management method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the various steps of the above-mentioned intelligent battery management method.

[0015] The above-mentioned intelligent battery management method, device, equipment and storage medium. In the embodiment of the present invention, by collecting the first multi-source original sensor parameter data corresponding to the first sensor parameter of the target intelligent battery and the second multi-source original sensor parameter data corresponding to the second sensor parameter, and performing spatiotemporal correlation and multiple preprocessing on the first multi-source original sensor parameter data and the second multi-source original sensor parameter data respectively, the first multi-modal sensor data and the second multi-modal sensor data are obtained; multi-level combined features are extracted from the first multi-modal sensor data and the second multi-modal sensor data to obtain a multi-battery parameter combination feature set, and multi-dimensional feature fusion is performed on each battery parameter combination feature set to obtain a fused feature vector of the intelligent battery state; the fused feature vector is subjected to the first multi-source sensor parameter pass The first channel and the second multi-source sensing parameter channel are separated to obtain the electrochemical state characteristic flow and the structural evolution characteristic flow, and the dual-channel cross-attention weights are calculated and the multi-target timing is predicted for the electrochemical state characteristic flow and the structural evolution characteristic flow to obtain the multi-dimensional state evaluation index of the target smart battery; the timing anomaly detection and the abnormal spatial distribution are determined for the multi-dimensional state evaluation index to obtain the multi-dimensional fault feature vector, and the propagation path calculation and risk level division are performed on the multi-dimensional fault feature vector to obtain the fault risk level of the target smart battery; the safety threshold grading calculation and the conversion of management instructions are performed on the fault risk level and the multi-dimensional state evaluation index to generate the battery management result of the target smart battery. Compared with the prior art, this application acquires basic state characteristics by multi-source data collection and feature separation of traditional electrochemical parameters and non-traditional physical parameters of the target smart battery, and then performs frequency domain decomposition and spatiotemporal reconstruction of these characteristics to obtain multimodal data of the battery; then, through multi-level combined feature extraction and fusion, the deep coupling of electrochemical characteristics and structural characteristics is realized to obtain fused feature vectors; then, based on the dual-channel cross-attention mechanism and multi-target time series prediction, the comprehensive state evaluation index of the battery is obtained, and the battery management results of the target smart battery are generated based on the comprehensive state evaluation index. Through hierarchical feature extraction and state evaluation, the problem of accurate identification of battery state monitoring is solved, especially in terms of performance degradation, structural evolution and fault diagnosis, the dynamic characteristics and degradation mechanism of the battery are fully considered, and the evaluation accuracy is effectively improved; and the multi-level feature fusion and risk warning strategy are adopted to realize the collaborative analysis of multi-source parameters and enhance the reliability of fault diagnosis; in addition, through fault propagation analysis and management instruction optimization, intelligent management and safety control of battery packs are realized.

[0016] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of a first embodiment of an intelligent battery management method according to an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of an intelligent battery management device in an embodiment of the present invention; Figure 3 FIG. 1 is a schematic diagram of an embodiment of an intelligent battery management device in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products, or device ends.

[0021] To facilitate understanding of this embodiment, the specific process of the embodiment of the present invention is described below. Figure 1 , the first embodiment of the intelligent battery management method in the embodiment of the present invention includes: 101. Collecting first multi-source original sensor parameter data corresponding to a first sensor parameter and second multi-source original sensor parameter data corresponding to a second sensor parameter of a target smart battery, and performing spatiotemporal correlation and multiple preprocessing on the first multi-source original sensor parameter data and the second multi-source original sensor parameter data, respectively, to obtain first multimodal sensor data and second multimodal sensor data; The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0022] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0023] In this embodiment, the first sensing parameter is a basic parameter for monitoring the working state of the battery (such as voltage parameter, current parameter, temperature parameter and internal resistance parameter), which mainly reflects the electrochemical performance and thermodynamic characteristics of the battery; the second sensing parameter refers to a structural response parameter (vibration parameter, acoustic emission parameter, ultrasonic parameter and displacement parameter), which mainly reflects the mechanical structure change of the battery. In addition, the first sensing parameter and the second sensing parameter can also be appropriately increased or decreased according to actual needs; the battery voltage data, battery current data, battery temperature data and battery internal resistance data corresponding to the first sensing parameter of the target smart battery are collected respectively, and based on the battery voltage data, the battery current data, the battery temperature data and the battery internal resistance data, the first multi-source original sensing parameter data is generated; the battery vibration data, battery acoustic emission data, battery ultrasonic data and battery displacement data corresponding to the second sensing parameter of the target smart battery are collected respectively, and based on the battery vibration data, the battery acoustic emission data, the battery ultrasonic data and the battery displacement data, Generate second multi-source original sensor parameter data; perform frequency segmentation on the first multi-source original sensor parameter data to obtain electrochemical low-frequency data segments, electrochemical medium-frequency data segments and electrochemical high-frequency data segments, and perform frequency segmentation on the second multi-source original sensor parameter data to obtain mechanical low-frequency data segments, mechanical medium-frequency data segments and mechanical high-frequency data segments; perform Kalman filtering on the electrochemical low-frequency data segments and the mechanical low-frequency data segments to obtain state basic characteristics, and perform wavelet multi-scale decomposition on the electrochemical medium-frequency data segments and the mechanical medium-frequency data segments to obtain dynamic response characteristics, and perform empirical mode decomposition on the electrochemical high-frequency data segments and the mechanical high-frequency data segments to obtain transient evolution characteristics; perform spatiotemporal synchronous matching on the state basic characteristics, the dynamic response characteristics and the transient evolution characteristics to obtain initial multimodal sensor data, and perform signal-to-noise ratio evaluation and weighted calculation on the initial multimodal sensor data corresponding to the original sensor parameters to obtain first multimodal sensor data and second multimodal sensor data.

[0024] In practical applications, the traditional electrochemical parameter data of the battery corresponding to the first sensing parameter (i.e., the first multi-source original sensing parameter data) is first collected through a high-precision sensor network, including the use of a voltage sensor with a sampling rate of 1kHz and an accuracy of 0.1mV to collect battery voltage data, a Hall current sensor with a sampling rate of 1kHz and an accuracy of 0.1% to collect battery current data, a distributed PT100 temperature sensor with a sampling rate of 100Hz to collect temperature field data, and an AC impedance tester with a sweep frequency in the frequency range of 0.1Hz-1kHz to collect internal resistance data; at the same time, a multi-source physical sensor is used The structural response parameters of the battery corresponding to the second sensing parameters (i.e., the second multi-source original sensing parameter data) are collected, including using a three-axis acceleration sensor to collect vibration data at a sampling rate of 10kHz, using a piezoelectric sensor array to collect acoustic emission signals in the range of 100kHz-1MHz, and using an ultrasonic transducer array to collect ultrasonic data in the frequency band of 20-100kHz, and using a fiber grating sensor to collect displacement data at a sampling rate of 50Hz and an accuracy of 1μm; and then the collected traditional electrochemical parameter data and structural response parameter data are time-space aligned and frequency segmented, that is, based on the spatial arrangement and response characteristics of the sensors, a data set is established. The time-space mapping relationship of the data is used to achieve synchronous alignment of multi-source data, and the electrochemical parameter data are divided into low-frequency band (0-100Hz), medium-frequency band (100Hz-10kHz) and high-frequency band (>10kHz) according to the frequency characteristics. Similarly, the structural response parameter data are also divided into three corresponding frequency bands. For the low-frequency data segment, the improved Kalman filter algorithm is used to eliminate the slowly varying noise and system drift, and the steady-state characteristics of the battery are extracted; for the medium-frequency data segment, the db4 wavelet basis function is used to perform a 5-layer multi-scale decomposition to separate the dynamic response characteristics of the battery; for the high-frequency data segment, the empirical mode decomposition method is used to extract the transient characteristics and nonlinear response characteristics; then the characteristics obtained by processing in different frequency bands are matched and fused in time and space synchronously, that is, by establishing a time reference and a spatial registration relationship, the basic state characteristics, dynamic response characteristics and transient evolution characteristics are aligned and integrated to form the initial multimodal sensing data; then the signal-to-noise ratio of the integrated data is evaluated, and the data quality indicators of each sensing parameter are calculated, including signal integrity, consistency and validity, etc., and the data are weighted based on these indicators, so that through this hierarchical data processing and feature extraction method, high-quality first multimodal sensing data and second multimodal sensing data are finally obtained.For example, during the charge and discharge process of lithium-ion batteries at a rate of 0.5C, the characteristic data obtained through the above processing show that the voltage-current coupling characteristics reflect the change of the charge state in the low frequency band (<100Hz), the polarization process in the medium frequency band, and the microscopic electrochemical reaction in the high frequency band; at the same time, the acoustic emission signal detects the formation of microcracks in the electrode material in the 50-200kHz frequency band, and the modal analysis of the vibration signal reveals the stress distribution of the internal structure of the battery. These characteristics form a multi-modal mapping relationship with the electrochemical parameters, providing multi-dimensional feature support for subsequent status evaluation.

[0025] 102. Extract multi-level combined features from the first multimodal sensing data and the second multimodal sensing data to obtain a multi-battery parameter combined feature set, and perform multi-dimensional feature fusion on each battery parameter combined feature set to obtain a fused feature vector of the intelligent battery state; In this embodiment, hierarchical feature extraction is performed on the first multimodal sensing data to obtain a voltage and current feature set and a temperature internal resistance feature set, and hierarchical feature extraction is performed on the second multimodal sensing data to obtain a vibration acoustic emission feature set and an ultrasonic displacement feature set; the voltage and current feature set and the vibration acoustic emission feature set are time-series paired to obtain a multidimensional feature relationship matrix, and the multidimensional feature relationship matrix is ​​cross-correlation calculated and nonlinearly mapped to obtain a first battery parameter combination feature set; the temperature internal resistance feature set and the ultrasonic displacement feature set are spatially aligned to obtain a structural parameter relationship matrix, and the structural parameter relationship matrix is ​​time-frequency mapped to obtain a second battery parameter combination feature set; the first battery parameter The combined feature set is subjected to multi-level wavelet packet decomposition to obtain a multi-scale battery reaction stress feature, and the second battery parameter combined feature set is subjected to empirical mode decomposition to obtain a multi-scale thermal field deformation feature, and the multi-scale battery reaction stress feature is subjected to energy reconstruction and feature selection to obtain an electrochemical strain coupling feature, and the multi-scale thermal field deformation feature is subjected to modal screening and feature fusion to obtain a temperature deformation coupling feature; hierarchical contribution calculation is performed on the electrochemical strain coupling feature and the temperature deformation coupling feature to obtain a multiple coupling weight matrix, and based on the multiple coupling weight matrix, the electrochemical strain coupling feature and the temperature deformation coupling feature are weightedly fused to obtain a fused feature vector of the smart battery state.

[0026] In practical applications, the first multimodal sensing data is firstly analyzed in layers, that is, the charging and discharging characteristics, polarization characteristics and capacity characteristics are extracted through voltage-current curve analysis, and the temperature field distribution and internal resistance change are correlated and analyzed to extract the thermal-resistance coupling characteristics, thereby obtaining the voltage and current feature set and the temperature internal resistance feature set; at the same time, the second multimodal sensing data is subjected to feature extraction, and the vibration signal and the acoustic emission signal are subjected to time-frequency analysis to extract the modal characteristics and energy characteristics, and the ultrasonic propagation characteristics and the displacement field distribution are subjected to spatial correlation analysis to extract the structural deformation characteristics to form the vibration acoustic emission feature set and the ultrasonic displacement feature set; the voltage and current feature set and the vibration acoustic emission feature set are then paired according to the time series, and the mapping relationship between the electrochemical response and the mechanical response is established by calculating the correlation coefficient and the time delay parameter between the features, and a multi-dimensional feature relationship matrix is ​​constructed, and the matrix is ​​subjected to nonlinear mapping and cross-correlation analysis to extract the electrochemical-mechanical coupling characteristics to obtain the first battery parameter combination feature set; similarly, the temperature internal resistance feature set and the ultrasonic displacement feature set are spatially registered to establish the corresponding relationship between the temperature field distribution and the structural deformation to form The structural parameter relationship matrix is ​​obtained, and then the second battery parameter combination feature set is obtained through time-frequency domain feature mapping; the first battery parameter combination feature set is then decomposed by multi-level wavelet packet transform to extract battery reaction-stress characteristics in different frequency bands and capture the coupling relationship between electrochemical reaction and mechanical stress; at the same time, the second battery parameter combination feature set is subjected to empirical mode decomposition to obtain multi-scale features reflecting temperature field distribution and structural deformation, and then the battery reaction-stress characteristics are integrated into electrochemical strain coupling characteristics through energy reconstruction and feature selection, and the thermal field-deformation characteristics are integrated into temperature deformation coupling characteristics through mode screening and feature fusion; a multi-level feature importance evaluation method is then used to calculate the contribution of electrochemical strain coupling characteristics and temperature deformation coupling characteristics to the battery state evolution, that is, the SHAP value of the feature is calculated by the XGBoost algorithm, combined with the feature screening results of Lasso regression and the feature importance score of random forest to construct a multi-coupling weight matrix, and based on the weight matrix, the electrochemical strain coupling characteristics and temperature deformation coupling characteristics are weightedly fused to finally obtain a fused feature vector that fully reflects the battery state. This multi-level feature extraction and fusion method not only achieves deep coupling of battery electrochemical and structural characteristics, but also can accurately capture the performance degradation and fault evolution characteristics of the battery. For example, the correlation analysis of the voltage drop and vibration signal of lithium-ion batteries shows that the sharp drop in voltage at the end of discharge is often accompanied by a significant change in the characteristic frequency. This correlation feature can be used to warn of the risk of battery failure in advance.

[0027] 103. Separate the fused feature vector into a first multi-source sensing parameter channel and a second multi-source sensing parameter channel to obtain an electrochemical state feature flow and a structural evolution feature flow, and perform dual-channel cross-attention weight calculation and multi-target time series prediction on the electrochemical state feature flow and the structural evolution feature flow to obtain a multi-dimensional state evaluation index of the target smart battery; In this embodiment, the attention weight of the fused feature vector corresponding to the first multi-source sensing parameter is calculated, the initial electrochemical feature is extracted, and the feature dimension of the initial electrochemical feature is compressed and the electrochemical feature is reconstructed to obtain the electrochemical feature subset of the target smart battery; the electrochemical feature subset is extracted by double-branch temporal convolution feature to obtain an electrochemical response matrix, and the electrochemical response matrix is ​​reconstructed in time and the performance characteristics are enhanced to obtain an electrochemical state characteristic flow; the local similarity of the fused feature vector corresponding to the second multi-source sensing parameter is calculated, the initial structural feature of the corresponding characteristic component is extracted, and the structural feature of the initial structural feature is reconstructed and the key structural features are screened to obtain the structural evolution feature subset of the target smart battery; the structural evolution feature subset is extracted by double-branch spatial convolution feature to obtain a structural response matrix, and the structural response matrix is Spatial reconstruction and feature enhancement are performed to obtain a structural evolution feature flow; self-attention weight calculation in the time dimension is performed on the corresponding charge and discharge feature branches and polarization feature branches in the electrochemical state feature flow to obtain electrochemical enhancement features, and self-attention weight calculation in the space dimension is performed on the corresponding stress feature branches and deformation feature branches in the structural evolution feature flow to obtain structural enhancement features; dual-channel cross-attention mapping calculation is performed on the electrochemical enhancement features and the structural enhancement features to obtain a feature weight mapping matrix, and the feature weight mapping matrix is ​​subjected to sliding time series segmentation and battery performance trend prediction to obtain a state evolution sequence; based on a preset branch evaluation model, the state evolution sequence is evaluated for multi-branch states to obtain an initial evaluation value of the battery state, and the initial evaluation value of the battery state is calibrated with confidence and optimized indices to obtain a multi-dimensional state evaluation index of the target smart battery. The above-mentioned preset branch evaluation model refers to a dedicated model structure used to evaluate the three key states of the battery: namely, the state of charge (SOC) evaluation branch: based on the fusion model of the improved Auger Coulomb counting method and the Kalman filter, combined with the voltage-current characteristics and temperature compensation, to evaluate the instantaneous charging state of the battery; the state of health (SOH) evaluation branch: based on the comprehensive evaluation model of capacity decay rate, internal resistance growth rate and coulomb efficiency, the parameters are continuously optimized through online learning to evaluate the health of the battery; the remaining service life (RUL) prediction branch: based on the life prediction model of support vector regression, combined with historical degradation data and current state characteristics, predicts the remaining service life of the battery; these three branch models are optimized using historical data during the training phase to form a preset evaluation benchmark for real-time evaluation of the battery status.

[0028] In practical applications, the self-attention weights are first calculated for the features related to the first multi-source sensing parameters in the fused feature vector. By setting the query matrix and the key value matrix, the multi-head self-attention mechanism is used to calculate the correlation scores between the features. The calculation formula of the multi-head self-attention mechanism is: , , , ; in, is the feature dimension, , , is a learnable weight matrix, F is a fused feature vector [n×d], n represents the number of time points of data, d contains all fused features, Q, K, V are query, key, and value matrices [n× ], softmax is a normalization function (for example: for the voltage feature vector [1.2V, 3.6V, 4.2V], the attention score [0.2, 0.3, 0.5] is calculated to indicate the importance of different voltage points), thereby extracting the initial electrochemical features reflecting the electrochemical performance of the battery; and using a variational autoencoder to reduce the dimension of these initial electrochemical features, the encoder maps the high-dimensional features to the latent space, and then the decoder reconstructs the electrochemical features to retain the key electrochemical information, thereby obtaining a subset of the electrochemical features of the battery (for example: compressing 50-dimensional electrochemical features to a 10-dimensional latent space, and then reconstructing the key electrochemical features); and then inputting the electrochemical feature subset into a dual-branch temporal convolutional network, in which one branch extracts the charge and discharge dynamic features through a one-dimensional convolution, and the other branch extracts the polarization dynamic features, wherein the dual-branch temporal convolutional network formula is: ; Among them, W(τ) is the convolution kernel, τ is the time delay, Xt is the timing feature, W is the convolution kernel parameter, Ht is the convolution output, τ is the time window (for example: use a convolution kernel of length 5 to extract features from the charge and discharge curve to capture the voltage change rate characteristics), and finally the outputs of the two branches are combined to form an electrochemical response matrix; then the response matrix is ​​reconstructed and feature enhanced in the timing dimension, and the attention mechanism is used to highlight the key timing features, and finally the state feature flow that characterizes the evolution of the battery's electrochemical performance is obtained. At the same time, the local sensitive hash is calculated for the second multi-source sensor parameter feature in the fusion feature vector, the local similarity between the features is evaluated, and the initial features reflecting the battery structure change are extracted. These initial structural features are reduced in dimension and reconstructed through the variational autoencoder, and the features most sensitive to the structural evolution are screened out to obtain the structural evolution feature subset of the battery; the structural evolution feature subset is then input into the dual-branch spatial convolutional network, one branch is responsible for extracting the stress distribution features, and the other branch is responsible for extracting the deformation distribution features. The outputs of the two branches are fused to obtain the structural response matrix; the spatial dimension of the response matrix is ​​then reconstructed and the features are enhanced, and the structural change features of the key areas are highlighted through the spatial attention mechanism, and the feature flow that characterizes the battery structure evolution law is obtained. Not only the effective decoupling of electrochemical and structural features is achieved, but also the integrity and representativeness of their respective features are maintained.

[0029] Secondly, the charge-discharge feature branch and polarization feature branch in the electrochemical state feature flow are processed in time series. By constructing a time series attention mechanism, the association weights of features at different time points are calculated, that is, the charge-discharge curve features are used as query sequences and the polarization features are used as key value sequences. Their self-attention scores in the time series dimension are calculated to highlight the electrochemical features of key time series nodes and obtain enhanced electrochemical feature expressions (for example, for the charge-discharge feature sequence [1.2V, 1.5V, 1.3V] and the polarization feature sequence [0.1Ω, 0.15Ω, 0.12Ω], enhanced features are obtained through time series self-attention calculation to highlight the electrochemical response at key moments). At the same time, the stress feature branch and the deformation feature branch in the structural evolution feature flow are calculated in the spatial dimension. By establishing a spatial attention mapping relationship, the importance weights of features at different spatial positions are calculated to highlight the structural change features of key areas and obtain enhanced structural feature expressions; the enhanced electrochemical features and structural features are then input into the dual-channel cross-attention network, and the cross-attention network formula is: ; in: , , , Fc is the electrochemical feature, Fs is the structural feature, Wq, Wk, Wv are projection matrices, d is the feature dimension, and the dot product Represents the correlation between electrochemical characteristics and structural characteristics, and thus establishes a mapping relationship between electrochemical state and structural evolution by calculating the mutual correlation between the two types of characteristics, forming a feature weight mapping matrix; then the matrix is ​​used for time series segmentation and trend prediction using the feature weight mapping matrix, and the appropriate window length and sliding step size are set to extract the time series characteristics of battery performance evolution, and these time series characteristics are modeled through a deep recurrent network to predict the change trend of battery performance. The time series prediction model corresponding to the feature weight mapping matrix is: ; Among them: X(t) is the electrochemical characteristic time series, Z(t) is the structural characteristic time series , is the basis function, , is the weight coefficient, τ is the prediction step length, and N is the number of features or basis functions (for example, by analyzing the capacity decay trend through a sliding window (window length 100, step length 10), predicting the health status of the next 50 cycles), so as to obtain a complete state evolution sequence; then based on the pre-trained three-branch evaluation model, the state of charge, health status and remaining life in the state evolution sequence are evaluated respectively to obtain the initial evaluation value of the battery state, that is, by establishing a confidence interval based on historical data statistics, the initial evaluation value is calibrated for reliability, and combined with the constraints of the battery physical model, the various state indicators are optimized, and finally a multi-dimensional evaluation indicator that accurately reflects the comprehensive state of the battery is obtained. It not only takes into account the interaction between electrochemical characteristics and structural characteristics, but also realizes the accurate evaluation of the multi-dimensional state of the battery.

[0030] 104. Perform time series anomaly detection and determine the abnormal spatial distribution of the multidimensional state evaluation index to obtain a multidimensional fault feature vector, and perform propagation path calculation and risk level division on the multidimensional fault feature vector to obtain the fault risk level of the target smart battery; In this embodiment, the deviation of the preset multi-monitoring parameters of the multi-dimensional state evaluation index is calculated to obtain a state change sequence, and the state change sequence is classified and abnormally located by multi-level battery abnormal working thresholds to obtain an abnormal state sequence; the abnormal state sequence is spatially mapped and associated features are extracted to obtain a multi-dimensional fault feature vector, and a fault association network is constructed for the multi-dimensional fault feature vector to obtain an initial propagation path matrix; the diffusion rate and propagation direction of the initial propagation path matrix are calculated to obtain fault dynamic evolution parameters, and the fault dynamic evolution parameters are reconstructed in time and space sequence to obtain a battery fault propagation matrix; the fault propagation matrix is ​​multi-dimensionally calculated for risk indicators to obtain a fault risk score, and the fault risk score is graded and prioritized to obtain the fault risk level of the target smart battery. The above-mentioned preset multi-monitoring parameters refer to the voltage change rate of the state of charge, the capacity decay rate of the health state, and the performance degradation rate of the remaining life.

[0031] In practical applications, reference values ​​of the normal working range of the battery are set for the multi-dimensional state evaluation indicators, including the voltage range of the state of charge, the capacity range of the healthy state, and the degradation threshold of the life prediction. The deviation between the actual state indicator and the reference value is calculated. The deviation calculation formula is: ; Among them, ΔS(t) is the comprehensive state deviation, is the weight coefficient of each state indicator, is the current state value, is the reference state value, is the time attenuation coefficient, is the initial moment, α is the spatial diffusion coefficient, and ∇² is the Laplace operator, which represents the spatial diffusion effect, thereby obtaining a state sequence that reflects the change in battery performance (for example, when the battery health state SOH drops from 90% to 85%, calculate its state deviation: ); then, based on the preset multi-level warning threshold, the state change sequence is detected for anomalies, including threshold judgments of three levels: slight anomaly, moderate anomaly and severe anomaly. At the same time, combined with the constraints of the battery physical model, the detected abnormal state is spatially located to determine the specific location and time point of the abnormality, thereby obtaining a complete abnormal state sequence; then, the abnormal state sequence is mapped to the spatial structure model of the battery, the distribution law and propagation characteristics of the abnormal state inside the battery are analyzed, the correlation characteristics between the abnormal states are extracted, a multi-dimensional fault feature vector is constructed, and a fault association network is established based on the feature vector. By calculating the mutual influence degree and propagation path between the fault points, an initial propagation network is formed. Path matrix; then quantitatively analyze the matrix, that is, calculate the diffusion rate and propagation direction of the fault inside the battery, obtain the key parameters describing the dynamic evolution of the fault, and integrate these parameters into a complete fault propagation matrix through the time-space sequence reconstruction method. The matrix fully reflects the development law and impact range of the fault; then calculate multidimensional risk indicators based on the fault propagation matrix, including the diffusion range, propagation speed and impact of the fault, and obtain the fault risk score through comprehensive evaluation. According to the distribution characteristics and clustering results of the score, the fault risk is divided into different levels, and the urgency and harmfulness of the fault are considered. Prioritize various types of faults, and finally obtain the fault risk level of the battery pack. This realizes the full process analysis from abnormal detection to risk assessment.

[0032] 105. Perform safety threshold classification calculation and management instruction conversion on the fault risk level and multi-dimensional status evaluation index to generate the battery management result of the target smart battery.

[0033] In this embodiment, the fault risk level and the multi-dimensional state evaluation index are mapped to a safety factor and converted to a working range to obtain a battery working constraint parameter; the battery working constraint parameter is converted to a charging and discharging parameter corresponding to a power limit coefficient to obtain a power management instruction, and the cooling channel flow distribution and heat dissipation power distribution calculation corresponding to a temperature threshold are performed on the battery working constraint parameter to obtain a temperature management instruction, and the active balancing circuit parameter calculation corresponding to the single cell state difference is performed on the battery working constraint parameter to obtain a balancing management instruction; the power management instruction, the temperature management instruction and the balancing management instruction are prioritized and the battery state is regulated to generate a battery management result for the target smart battery.

[0034] In practical applications, a safety margin mapping relationship is established based on the fault risk level and the multi-dimensional state evaluation index, the fault risk level is converted into a power limit coefficient, and the state evaluation index is converted into a working range boundary. The battery working constraint parameters are obtained through comprehensive calculation (wherein these constraint parameters not only take into account the safety boundary of the battery, but also include the target requirements for performance optimization, providing a basic basis for the subsequent generation of management instructions); and then in the process of generating management instructions, the power limit coefficient in the working constraint parameters is first converted into specific charging and discharging parameters, including the upper limit of the charging current, the discharge power threshold and the cut-off voltage, etc., to form a power management instruction; at the same time, based on the temperature threshold parameter, the flow distribution ratio of each cooling channel and the radiator are calculated. The power distribution scheme takes into account the uniformity of temperature field distribution and the key control of hot spots, and generates temperature management instructions. In addition, according to the state differences between single cells, the working parameters of the balancing circuit are calculated, including the balancing current size, balancing time and balancing path selection, and energy transfer is achieved through the DC-DC conversion circuit to form balancing management instructions. Then, the three types of management instructions are coordinated and prioritized. When multiple management requirements appear, safety-related instructions are executed first, followed by performance optimization-related instructions, and finally maintenance management instructions are executed. Then, through the reasonable arrangement of execution priorities and coordination between instructions, the safe and controllable operation of the battery pack is achieved, and finally a complete battery management result is generated. This hierarchical management strategy not only ensures the safe operation of the battery, but also optimizes and improves performance.

[0035] In the embodiment of the present invention, the basic state characteristics are obtained by multi-source data collection and feature separation of the traditional electrochemical parameters and non-traditional physical parameters of the target smart battery, and then the characteristics are decomposed in the frequency domain and reconstructed in time and space to obtain multi-modal data of the battery; then, through multi-level combined feature extraction and fusion, the deep coupling of electrochemical characteristics and structural characteristics is realized to obtain the fusion feature vector; then, based on the dual-channel cross-attention mechanism and multi-target time series prediction, the comprehensive state evaluation index of the battery is obtained, so as to generate the battery management result of the target smart battery based on the comprehensive state evaluation index. Through hierarchical feature extraction and state evaluation, the problem of accurate identification of battery state monitoring is solved, especially in terms of performance degradation, structural evolution and fault diagnosis, the dynamic characteristics and degradation mechanism of the battery are fully considered, and the evaluation accuracy is effectively improved; and the multi-level feature fusion and risk warning strategy are adopted to realize the collaborative analysis of multi-source parameters and enhance the reliability of fault diagnosis; in addition, through fault propagation analysis and management instruction optimization, the intelligent management and safety control of the battery pack are realized.

[0036] The above describes the intelligent battery management method in the embodiment of the present invention. The following describes the intelligent battery management device in the embodiment of the present invention. Figure 2, an embodiment of the intelligent battery management device in the embodiment of the present invention includes: A preprocessing module 201 is used to collect first multi-source original sensor parameter data corresponding to a first sensor parameter and second multi-source original sensor parameter data corresponding to a second sensor parameter of a target smart battery, and respectively perform spatiotemporal correlation and multiple preprocessing on the first multi-source original sensor parameter data and the second multi-source original sensor parameter data to obtain first multimodal sensor data and second multimodal sensor data; A feature fusion module 202 is used to extract multi-level combined features from the first multimodal sensing data and the second multimodal sensing data to obtain a multi-battery parameter combination feature set, and perform multi-dimensional feature fusion on each of the battery parameter combination feature sets to obtain a fused feature vector of the smart battery state; An index evaluation module 203 is used to separate the first multi-source sensing parameter channel and the second multi-source sensing parameter channel of the fused feature vector to obtain an electrochemical state feature flow and a structural evolution feature flow, and to calculate dual-channel cross-attention weights and predict multi-target time series on the electrochemical state feature flow and the structural evolution feature flow to obtain a multi-dimensional state evaluation index of the target smart battery; The risk classification module 204 is used to perform time series anomaly detection and anomaly spatial distribution determination on the multi-dimensional state evaluation index to obtain a multi-dimensional fault feature vector, and perform propagation path calculation and risk level classification on the multi-dimensional fault feature vector to obtain a fault risk level of the target smart battery; The instruction conversion module 205 is used to perform safety threshold classification calculation and management instruction conversion on the fault risk level and the multi-dimensional state evaluation index to generate a battery management result of the target smart battery.

[0037] In the embodiment of the present invention, the basic state characteristics are obtained by multi-source data collection and feature separation of the traditional electrochemical parameters and non-traditional physical parameters of the target smart battery, and then the characteristics are decomposed in the frequency domain and reconstructed in time and space to obtain multi-modal data of the battery; then, through multi-level combined feature extraction and fusion, the deep coupling of electrochemical characteristics and structural characteristics is realized to obtain the fusion feature vector; then, based on the dual-channel cross-attention mechanism and multi-target time series prediction, the comprehensive state evaluation index of the battery is obtained, so as to generate the battery management result of the target smart battery based on the comprehensive state evaluation index. Through hierarchical feature extraction and state evaluation, the problem of accurate identification of battery state monitoring is solved, especially in terms of performance degradation, structural evolution and fault diagnosis, the dynamic characteristics and degradation mechanism of the battery are fully considered, and the evaluation accuracy is effectively improved; and the multi-level feature fusion and risk warning strategy are adopted to realize the collaborative analysis of multi-source parameters and enhance the reliability of fault diagnosis; in addition, through fault propagation analysis and management instruction optimization, the intelligent management and safety control of the battery pack are realized.

[0038] above Figure 2 The intelligent battery management device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The intelligent battery management device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0039] Figure 3 3 is a schematic diagram of the structure of an intelligent battery management device provided by an embodiment of the present invention. The intelligent battery management device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage devices) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the intelligent battery management device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330 to execute a series of instruction operations in the storage medium 330 on the intelligent battery management device 300.

[0040] The intelligent battery management device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the intelligent battery management device shown does not constitute a limitation on the intelligent battery management device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0041] The present invention also provides an intelligent battery management device, wherein the computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes each step of the intelligent battery management method in the above-mentioned embodiments.

[0042] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the various steps of the intelligent battery management method.

[0043] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0044] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0045] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0046] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent battery management method, characterized in that: The intelligent battery management method comprises: Collecting first multi-source original sensor parameter data corresponding to a first sensor parameter and second multi-source original sensor parameter data corresponding to a second sensor parameter of a target smart battery, and performing spatiotemporal correlation and multiple preprocessing on the first multi-source original sensor parameter data and the second multi-source original sensor parameter data, respectively, to obtain first multimodal sensor data and second multimodal sensor data; Extracting multi-level combined features from the first multimodal sensing data and the second multimodal sensing data to obtain a multi-battery parameter combined feature set, and performing multi-dimensional feature fusion on each of the battery parameter combined feature sets to obtain a fused feature vector of the intelligent battery state; Separating the fused feature vector into a first multi-source sensing parameter channel and a second multi-source sensing parameter channel to obtain an electrochemical state feature flow and a structural evolution feature flow, and performing dual-channel cross-attention weight calculation and multi-target time series prediction on the electrochemical state feature flow and the structural evolution feature flow to obtain a multi-dimensional state evaluation index of the target smart battery; Performing time series anomaly detection and determining anomaly spatial distribution on the multidimensional state evaluation index to obtain a multidimensional fault feature vector, and performing propagation path calculation and risk level division on the multidimensional fault feature vector to obtain a fault risk level of the target smart battery; The fault risk level and the multi-dimensional state evaluation index are subjected to safety threshold classification calculation and management instruction conversion to generate a battery management result of the target smart battery.

2. The intelligent battery management method according to claim 1, characterized in that: The step of performing spatiotemporal correlation and multiple preprocessing on the first multi-source original sensor parameter data and the second multi-source original sensor parameter data to obtain first multimodal sensor data and second multimodal sensor data includes: Frequency segmenting the first multi-source original sensing parameter data to obtain electrochemical low-frequency data segments, electrochemical medium-frequency data segments, and electrochemical high-frequency data segments, and frequency segmenting the second multi-source original sensing parameter data to obtain mechanical low-frequency data segments, mechanical medium-frequency data segments, and mechanical high-frequency data segments; Performing Kalman filtering on the electrochemical low-frequency data segment and the mechanical low-frequency data segment to obtain state basic characteristics, performing wavelet multi-scale decomposition on the electrochemical intermediate-frequency data segment and the mechanical intermediate-frequency data segment to obtain dynamic response characteristics, and performing empirical mode decomposition on the electrochemical high-frequency data segment and the mechanical high-frequency data segment to obtain transient evolution characteristics; The state basic characteristics, the dynamic response characteristics and the transient evolution characteristics are synchronously matched in time and space to obtain initial multimodal sensing data, and the signal-to-noise ratio of the corresponding original sensing parameters is evaluated and weighted calculated on the initial multimodal sensing data to obtain first multimodal sensing data and second multimodal sensing data.

3. The intelligent battery management method according to claim 1, characterized in that: The extracting of multi-level combined features of the first multimodal sensing data and the second multimodal sensing data to obtain a multi-battery parameter combined feature set, and performing multi-dimensional feature fusion on each of the battery parameter combined feature sets to obtain a fused feature vector of the intelligent battery state includes: Performing hierarchical feature extraction on the first multimodal sensing data to obtain a voltage and current feature set and a temperature internal resistance feature set, and performing hierarchical feature extraction on the second multimodal sensing data to obtain a vibration acoustic emission feature set and an ultrasonic displacement feature set; Performing time-series pairing on the voltage and current feature set and the vibration and acoustic emission feature set to obtain a multi-dimensional feature relationship matrix, and performing cross-correlation calculation and nonlinear mapping on the multi-dimensional feature relationship matrix to obtain a first battery parameter combination feature set; Performing spatial registration on the temperature internal resistance feature set and the ultrasonic displacement feature set to obtain a structural parameter relationship matrix, and performing time-frequency feature mapping on the structural parameter relationship matrix to obtain a second battery parameter combination feature set; Performing wavelet packet multi-level decomposition on the first battery parameter combination feature set to obtain a multi-scale battery reaction stress feature, performing empirical mode decomposition on the second battery parameter combination feature set to obtain a multi-scale thermal field deformation feature, performing energy reconstruction and feature selection on the multi-scale battery reaction stress feature to obtain an electrochemical strain coupling feature, and performing modal screening and feature fusion on the multi-scale thermal field deformation feature to obtain a temperature deformation coupling feature; The electrochemical strain coupling feature and the temperature deformation coupling feature are hierarchically calculated for contribution to obtain a multiple coupling weight matrix, and based on the multiple coupling weight matrix, the electrochemical strain coupling feature and the temperature deformation coupling feature are weightedly fused to obtain a fused feature vector of the smart battery state.

4. The intelligent battery management method according to claim 1, characterized in that: The step of separating the first multi-source sensing parameter channel and the second multi-source sensing parameter channel of the fused feature vector to obtain an electrochemical state feature flow and a structural evolution feature flow includes: Calculating the attention weight of the fused feature vector corresponding to the first multi-source sensing parameter, extracting the initial electrochemical feature, and compressing the feature dimension and reconstructing the electrochemical feature of the initial electrochemical feature to obtain an electrochemical feature subset of the target smart battery; Extracting double-branch time-series convolution features from the electrochemical feature subset to obtain an electrochemical response matrix, and performing time-series reconstruction and performance feature enhancement on the electrochemical response matrix to obtain an electrochemical state feature flow; Calculating the local similarity of the fused feature vector corresponding to the second multi-source sensing parameter, extracting the initial structural features of the corresponding feature components, and reconstructing the structural features and screening the key structural features of the initial structural features to obtain a structural evolution feature subset of the target smart battery; A dual-branch spatial convolution feature extraction is performed on the structural evolution feature subset to obtain a structural response matrix, and the structural response matrix is ​​spatially reconstructed and feature enhanced to obtain a structural evolution feature flow.

5. The intelligent battery management method according to claim 1, characterized in that: The calculation of the dual-channel cross-attention weights and the prediction of the multi-target time series of the electrochemical state characteristic flow and the structural evolution characteristic flow are performed to obtain the multi-dimensional state evaluation index of the target smart battery, including: Performing self-attention weight calculation in the time dimension on the corresponding charge-discharge feature branch and polarization feature branch in the electrochemical state feature flow to obtain electrochemical enhancement features, and performing self-attention weight calculation in the space dimension on the corresponding stress feature branch and deformation feature branch in the structural evolution feature flow to obtain structural enhancement features; Performing dual-channel cross-attention mapping calculation on the electrochemical enhancement feature and the structural enhancement feature to obtain a feature weight mapping matrix, and performing sliding time series segmentation and battery performance trend prediction on the feature weight mapping matrix to obtain a state evolution sequence; Based on a preset branch evaluation model, the state evolution sequence is evaluated for multi-branch states to obtain an initial evaluation value of the battery state, and the initial evaluation value of the battery state is calibrated for confidence and optimized for indicators to obtain a multi-dimensional state evaluation indicator of the target smart battery.

6. The intelligent battery management method according to claim 1, characterized in that: The multidimensional state evaluation index is subjected to time series anomaly detection and determination of anomaly spatial distribution to obtain a multidimensional fault feature vector, and the multidimensional fault feature vector is subjected to propagation path calculation and risk level division to obtain the fault risk level of the target smart battery, including: Performing deviation calculation of preset multiple monitoring parameters on the multi-dimensional state evaluation index to obtain a state change sequence, and performing classification and abnormal location of multi-level battery abnormal working thresholds on the state change sequence to obtain an abnormal state sequence; Performing spatial distribution mapping and correlation feature extraction on the abnormal state sequence to obtain a multi-dimensional fault feature vector, and constructing a fault correlation network on the multi-dimensional fault feature vector to obtain an initial propagation path matrix; Calculating the diffusion rate and propagation direction of the initial propagation path matrix to obtain fault dynamic evolution parameters, and reconstructing the fault dynamic evolution parameters in a spatiotemporal sequence to obtain a battery fault propagation matrix; A multi-dimensional risk index calculation is performed on the fault propagation matrix to obtain a fault risk score, and the fault risk score is graded and prioritized to obtain a fault risk level of the target smart battery.

7. The intelligent battery management method according to claim 1, characterized in that: The step of performing safety threshold classification calculation and management instruction conversion on the fault risk level and the multi-dimensional state evaluation index to generate a battery management result of the target smart battery includes: Performing safety factor mapping and working range conversion on the fault risk level and the multi-dimensional state evaluation index to obtain battery working constraint parameters; Converting the battery operation constraint parameters into charge and discharge parameters corresponding to the power limit coefficient to obtain a power management instruction, calculating the cooling channel flow distribution and heat dissipation power distribution corresponding to the temperature threshold to obtain a temperature management instruction, and calculating the active balancing circuit parameters corresponding to the single cell state difference to obtain a balancing management instruction; The power management instructions, the temperature management instructions and the balancing management instructions are prioritized and the battery status is regulated to generate a battery management result of the target smart battery.

8. An intelligent battery management device, characterized in that: The intelligent battery management device comprises: A preprocessing module, used for collecting first multi-source original sensor parameter data corresponding to the first sensor parameter and second multi-source original sensor parameter data corresponding to the second sensor parameter of the target smart battery, and performing spatiotemporal correlation and multiple preprocessing on the first multi-source original sensor parameter data and the second multi-source original sensor parameter data, respectively, to obtain first multi-modal sensor data and second multi-modal sensor data; A feature fusion module, used to extract multi-level combined features from the first multimodal sensing data and the second multimodal sensing data to obtain a multi-battery parameter combination feature set, and perform multi-dimensional feature fusion on each of the battery parameter combination feature sets to obtain a fused feature vector of the intelligent battery state; An index evaluation module is used to separate the first multi-source sensing parameter channel and the second multi-source sensing parameter channel of the fused feature vector to obtain an electrochemical state feature flow and a structural evolution feature flow, and to calculate dual-channel cross-attention weights and predict multi-target time series on the electrochemical state feature flow and the structural evolution feature flow to obtain a multi-dimensional state evaluation index of the target smart battery; A risk classification module, used to perform time series anomaly detection and anomaly spatial distribution determination on the multi-dimensional state evaluation index to obtain a multi-dimensional fault feature vector, and perform propagation path calculation and risk level classification on the multi-dimensional fault feature vector to obtain a fault risk level of the target smart battery; The instruction conversion module is used to perform safety threshold classification calculation and management instruction conversion on the fault risk level and the multi-dimensional state evaluation index to generate a battery management result of the target smart battery.

9. An intelligent battery management device, characterized in that: The intelligent battery management device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the intelligent battery management device to perform each step of the intelligent battery management method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the intelligent battery management method as described in any one of claims 1 to 7 are implemented.

Images