Method and system for detecting health state of vehicle battery based on multi-source data analysis

Abstract

The invention discloses a method and a system for detecting the health state of a vehicle battery based on multi-source data analysis, and relates to the technical field of big data analysis, and the method comprises the steps: collecting multi-source data and historical capacity attenuation data of the vehicle battery, carrying out the space-time correlation modeling of the multi-source data, generating a space-time correlation matrix, and carrying out the detection of the health state of the vehicle battery based on the space-time correlation matrix. The method comprises the following steps: extracting key features by adopting an attention mechanism, establishing a capacity fading prediction model based on the key features and historical capacity fading data, calculating a health state index of a current vehicle battery, judging whether the change rate of the health state index exceeds a preset fluctuation range or not, and if the change rate exceeds the preset fluctuation range, judging whether the change rate exceeds the preset fluctuation range. And if not, performing real-time processing on the newly collected multi-source data by adopting an incremental learning algorithm and updating model parameters of the capacity attenuation prediction model. And a reliable basis is provided for battery life prediction, fault diagnosis and maintenance decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis, and in particular to a method and system for detecting the health status of a vehicle battery based on multi-source data analysis. Background Art

[0002] As a core component of new energy vehicles, the health of automotive batteries is directly related to the vehicle's safety, reliability, and economy. During operation, automotive batteries experience complex operating environments and conditions, including varying temperatures, humidity, charge and discharge rates, and complex road conditions. These factors can cause battery performance to gradually degrade, resulting in a decrease in capacity and an increase in internal resistance, ultimately impacting the vehicle's range and service life.

[0003] Accurately assessing the health status of automotive batteries is crucial for preventing accidents, optimizing battery management systems, extending battery life, and improving user experience. However, battery aging is a complex, nonlinear process influenced by multiple factors, making it difficult for a single data source to fully reflect the battery's true health status. Therefore, intelligent health indicator analysis based on multi-source data, including voltage, current, temperature, internal resistance, and charge / discharge cycle count, is required to accurately assess the battery's health status and provide reliable decision-making support for the battery management system.

[0004] Existing automotive battery health status detection methods often use manually designed features in the feature extraction process, which makes it difficult to automatically identify and extract the most critical features for battery health status assessment, resulting in insufficient model generalization ability; and often ignore the time series characteristics and spatial correlation of the data, and cannot fully explore the deep information contained in the data.

[0005] To this end, the present invention proposes a method and system for detecting the health status of a vehicle battery based on multi-source data analysis. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a method and system for detecting the health status of automotive batteries based on multi-source data analysis, providing a reliable basis for battery life prediction, fault diagnosis, and maintenance decision-making.

[0007] To achieve the above objectives, a method for detecting the health status of a vehicle battery based on multi-source data analysis is proposed, comprising the following steps:

[0008] A method for detecting the health status of a vehicle battery based on multi-source data analysis includes the following steps:

[0009] Step 1: Collect multi-source data and historical capacity decay data of vehicle batteries;

[0010] Step 2: Perform spatiotemporal correlation modeling on the multi-source data to generate a spatiotemporal correlation matrix;

[0011] Step 3: Based on the spatiotemporal correlation matrix, an attention mechanism is used to extract key features, where the key features are used to characterize the nonlinear characteristics of the vehicle battery aging trajectory;

[0012] Step 4: Based on the key features and historical capacity decay data, a capacity decay prediction model is established to calculate the health status index of the current vehicle battery, and determine whether the rate of change of the health status index exceeds a preset fluctuation range. If so, proceed to step 5; otherwise, output the health status index;

[0013] Step 5: Use the incremental learning algorithm to process the newly collected multi-source data in real time and update the model parameters of the capacity decay prediction model, and then return to step 4.

[0014] The multi-source data of vehicle batteries is collected in the following manner:

[0015] In an automotive battery testing environment, a multi-source sensor array is used to form a sensor network for the automotive battery. The parameter values ​​of various physical parameters of the automotive battery are collected in real time through the sensor network, and the parameter values ​​of the various physical parameters are combined into the multi-source data. The physical parameters include voltage, current, temperature, internal resistance, and number of charge and discharge cycles.

[0016] The historical capacity decay data of the vehicle battery is collected as follows:

[0017] In an automotive battery test environment, while collecting multi-source data, the capacity of the automotive battery is tested. During each capacity test, the actual available capacity value of the complete charge and discharge process is recorded, and the corresponding relationship between the current cycle number and the actual available capacity value is recorded.

[0018] Generating the spatiotemporal correlation matrix comprises the following steps:

[0019] Step 21: Extract time series features from the collected multi-source data and construct time dimension features;

[0020] Step 22: Analyze the spatial correlation characteristics of the collected multi-source data and construct an interaction network between various physical parameters;

[0021] The spatial correlation analysis uses a graph neural network architecture, treating each physical parameter as a node in the graph neural network and the interactions between them as edges. The adjacency matrix A of the graph neural network is initialized to a fully connected form and then optimized through data-driven learning.

[0022] Step 23: Fuse the temporal dimension features and spatial correlation features to construct a spatiotemporal joint representation model;

[0023] The spatiotemporal joint representation model adopts a tensor decomposition architecture to organize time dimension features and spatial correlation features into a third-order tensor, and extracts core correlation patterns by decomposing the third-order tensor through Tucker.

[0024] Step 24: Based on the joint spatiotemporal representation, a multi-level spatiotemporal correlation matrix is ​​constructed to represent the parameter coupling relationships at different scales. The multi-level spatiotemporal correlation matrix is ​​derived from the Tucker decomposition results through a projection transformation. The multi-level spatiotemporal correlation matrix includes three levels of correlation matrices, representing the parameter coupling relationships at the micro, meso, and macro scales, respectively.

[0025] Step 25: weightedly fusing the multi-level correlation matrix based on preset weights to generate a final spatiotemporal correlation matrix to characterize the coupling effect between the driving condition and the charging condition of the vehicle battery;

[0026] The method of extracting key features using an attention mechanism based on the spatiotemporal correlation matrix includes the following steps:

[0027] Step 31: Construct a multi-head self-attention mechanism, and use the parallel attention calculation units in the multi-head self-attention mechanism to enhance the spatiotemporal correlation matrix and output a feature matrix.

[0028] Step 32: Based on the feature matrix output by the multi-head self-attention mechanism, a parameter-level attention mechanism is constructed to adaptively evaluate the importance of different physical parameters and identify key parameters that significantly affect the battery health status of the vehicle battery;

[0029] Step 33: Based on the key parameters output by the parameter-level attention mechanism, a time-scale attention mechanism is constructed by combining multi-scale temporal convolution with attention to capture battery aging characteristics at different time scales.

[0030] Step 34: Based on the battery aging characteristics output by the time-scale attention mechanism, a feature interaction attention mechanism is designed to model the nonlinear interaction relationship between various physical parameters;

[0031] Step 35: Fuse the output of the parameter-level attention mechanism, the output of the time-scale attention mechanism, and the output of the feature interaction attention mechanism to generate key features that characterize the nonlinear characteristics of the battery aging trajectory; the fusion of the key features adopts a gated feature fusion mechanism to adaptively integrate the durations of the parameter-level attention mechanism, the time-scale attention mechanism, and the feature interaction attention mechanism.

[0032] The establishing of a capacity decay prediction model based on the key features and historical capacity decay data comprises the following steps:

[0033] Step 41: Based on the extracted key features and historical capacity decay data, a multi-branch fusion network architecture is constructed; the multi-branch fusion network architecture includes three main components: a feature processing branch, a historical capacity branch, and a fusion prediction branch.

[0034] In the feature processing branch, a fully connected layer is constructed to perform feature transformation on the key features and output the changed key features;

[0035] Step 43: In the historical capacity branch, a recurrent neural network structure is constructed to capture the temporal dynamic characteristics of the historical capacity decay data; a temporal feature aggregation mechanism is designed to extract key information from the historical capacity decay data based on the temporal dynamic characteristics;

[0036] Step 44: By constructing the fusion prediction branch, the output of the feature processing branch and the output of the historical capacity branch are spliced ​​and fused to obtain a fusion feature;

[0037] Step 45: Based on the concatenated and fused features, a fully connected network is constructed to predict the capacity of the vehicle battery and calculate the health status indicator of the vehicle battery. The fully connected network includes two fully connected layers that map the fused features to the capacity prediction value. The first fully connected layer of the fully connected network is used for feature transformation, and the second fully connected layer of the fully connected network is used to generate the capacity prediction value.

[0038] Step 46: Based on the multi-branch fusion network architecture, a capacity decay prediction model is obtained by training the key features and the historical capacity decay data.

[0039] The method for calculating the health status indicator of the current vehicle battery is:

[0040] Calculating the health status indicators of the current vehicle battery using the capacity decay prediction model includes:

[0041] Inputting key characteristics and historical capacity decay data of the current vehicle battery into the capacity decay prediction model to obtain a predicted capacity value of the current vehicle battery;

[0042] Based on the predicted capacity value of the current vehicle battery, the health status index of the current vehicle battery is calculated using a predefined health status index calculation formula:

[0043] The method for determining whether the change rate of the health status indicator exceeds the preset fluctuation range is:

[0044] If the difference between two consecutive calculation results of the health status indicator is greater than a preset difference threshold, it is determined that the change rate of the health status indicator exceeds a preset fluctuation range;

[0045] The method of using an incremental learning algorithm to process newly collected data in real time and update the model parameters of the capacity decay prediction model includes the following steps:

[0046] Step 51: Extract features from the newly collected multi-source data to generate feature vectors for incremental learning;

[0047] Step 52: Construct a buffer to store the feature vectors generated and collected within a preset time and their corresponding actual capacity values;

[0048] Step 53: Based on the feature vectors and actual capacity values ​​stored in the buffer, the parameters of the capacity decay prediction model are updated using the gradient descent method;

[0049] A vehicle battery health status detection system based on multi-source data analysis is proposed, which includes a multi-source data collection module, a spatiotemporal matrix construction module, a key feature extraction module, a monitoring indicator generation module, and a model update module.

[0050] The multi-source data collection module collects multi-source data and historical capacity decay data of vehicle batteries, sends the multi-source data to the spatiotemporal matrix construction module, and sends the historical capacity decay data to the monitoring indicator generation module;

[0051] A spatiotemporal matrix construction module performs spatiotemporal correlation modeling on the multi-source data, generates a spatiotemporal correlation matrix, and sends the spatiotemporal correlation matrix to a key feature extraction module;

[0052] A key feature extraction module, which extracts key features based on the spatiotemporal correlation matrix using an attention mechanism and sends the key features to a monitoring indicator generation module;

[0053] A monitoring indicator generation module, which establishes a capacity decay prediction model based on the key characteristics and historical capacity decay data, calculates the health status indicator of the current vehicle battery, and sends the health status indicator to the model update module;

[0054] The model update module determines whether the rate of change of the health status indicator exceeds a preset fluctuation range. If so, an incremental learning algorithm is used to process the newly collected multi-source data in real time and update the model parameters of the capacity decay prediction model. Otherwise, the health status indicator is output.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1. By collecting multi-source data and historical capacity decay data for automotive batteries, and then performing spatiotemporal correlation modeling on the collected multi-source data, a spatiotemporal correlation matrix is ​​generated. This takes into account the time series characteristics of the data and the interrelationships between different parameters, and can more comprehensively capture the dynamic changes in the battery during operation;

[0057] 2. Based on the spatiotemporal correlation matrix, an attention mechanism is used to extract key features, automatically identifying and focusing on the data features that are most important for battery health assessment, improving the model's interpretability and accuracy.

[0058] 3. Based on the extracted key features and historical capacity decay data, a capacity decay prediction model is established to calculate the health status indicators of the current vehicle battery. When the rate of change of the health status indicators exceeds the preset fluctuation range, the system uses an incremental learning algorithm to process the newly collected multi-source data in real time, update the parameters of the capacity decay prediction model, and then recalculate the health status indicators. This adaptive update mechanism enables the model to continuously optimize as the battery ages, maintaining a high level of prediction accuracy. It enhances the intelligence level of the battery management system and provides a reliable basis for battery life prediction, fault diagnosis, and maintenance decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of a method for detecting the health status of a vehicle battery based on multi-source data analysis in Example 1 of the present invention;

[0060] Figure 2 This is an example diagram of the architecture of a vehicle battery health status detection system based on multi-source data analysis in Example 2 of the present invention. DETAILED DESCRIPTION

[0061] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0062] Example 1

[0063] like Figure 1 As shown, a method for detecting the health status of a vehicle battery based on multi-source data analysis includes the following steps:

[0064] Step 1: Collect multi-source data and historical capacity decay data of vehicle batteries;

[0065] Step 2: Perform spatiotemporal correlation modeling on the multi-source data to generate a spatiotemporal correlation matrix;

[0066] Step 3: Based on the spatiotemporal correlation matrix, an attention mechanism is used to extract key features, where the key features are used to characterize the nonlinear characteristics of the vehicle battery aging trajectory;

[0067] Step 4: Based on the key features and historical capacity decay data, a capacity decay prediction model is established to calculate the health status index of the current vehicle battery, and determine whether the rate of change of the health status index exceeds a preset fluctuation range. If so, proceed to step 5; otherwise, output the health status index;

[0068] Step 5: Use the incremental learning algorithm to process the newly collected multi-source data in real time and update the model parameters of the capacity decay prediction model, and then return to step 4.

[0069] In an embodiment of the present invention, the multi-source data of the vehicle battery is collected in the following manner:

[0070] In an automotive battery testing environment, a multi-source sensor array is used to form a sensor network for the automotive battery. The sensor network collects the parameter values ​​of various physical parameters of the automotive battery in real time. The parameter values ​​of various physical parameters constitute the multi-source data; the various physical parameters include voltage, current, temperature, internal resistance, and number of charge and discharge cycles.

[0071] Specifically, the vehicle battery test environment refers to a closed, controllable test system, in which the vehicle battery can be charged and discharged in a cycle test according to a preset test plan, while collecting and recording detailed operating data of the vehicle battery during the entire test process; the sensor network includes a voltage acquisition unit, a current acquisition unit, a temperature acquisition unit, an internal resistance measurement unit and a cycle counter.

[0072] Among them, the voltage acquisition unit adopts a high-precision analog-to-digital converter and realizes real-time acquisition of voltage data through digital signal processing technology; the current acquisition unit adopts a current sensor to realize bidirectional measurement of charging and discharging current; the temperature acquisition unit adopts a temperature sensor array and arranges multiple temperature measurement points on the surface of the vehicle battery to form a temperature field distribution monitoring network; the internal resistance measurement unit is based on impedance measurement technology to realize dynamic monitoring of the internal resistance characteristics of the vehicle battery.

[0073] The number of charge and discharge cycles is recorded by a dedicated cycle counter, which is used to identify complete charge and discharge cycles and partial charge and discharge processes, and performs weighted counting according to different depths.

[0074] In a further embodiment of the present invention, during the multi-source data acquisition process, each acquisition unit and cycle counter operates in a synchronous sampling mode. The voltage acquisition unit monitors the vehicle battery terminal voltage in real time, while simultaneously oversampling and digitally filtering the acquired data to suppress interference and noise in the working environment. The current acquisition unit ensures measurement accuracy through temperature compensation and zero-point calibration techniques. The temperature acquisition unit dynamically monitors the surface temperature distribution of the vehicle battery through multi-point temperature measurement, and the acquired data undergoes digital filtering and temperature calibration. The internal resistance measurement unit obtains the impedance characteristics of the vehicle battery through small-signal excitation when the vehicle battery is in a static state.

[0075] All collected multi-source data uses a unified time base and is transmitted via a data transmission network to the data processing backend to ensure time synchronization. The data processing backend performs preprocessing on the collected real-time multi-source data, including outlier detection, data smoothing, and feature extraction. Furthermore, the multi-source data is ultimately stored in a structured format, including information such as timestamps, parameter values, status identifiers, and quality indicators.

[0076] Furthermore, the historical capacity attenuation data of the vehicle battery is collected in the following manner:

[0077] In an automotive battery test environment, while collecting multi-source data, the capacity of the automotive battery is tested. During each capacity test, the actual available capacity value of the complete charge and discharge process is recorded, and the corresponding relationship between the current cycle number and the actual available capacity value is recorded.

[0078] Specifically, the acquired historical capacity decay data includes information such as the number of cycles, actual capacity value, and test timestamp.

[0079] Furthermore, generating the spatiotemporal correlation matrix includes the following steps:

[0080] Step 21: Extract time series features from the collected multi-source data and construct time dimension features.

[0081] In some embodiments, temporal feature extraction utilizes a multi-scale temporal convolutional network structure to process chronologically ordered time series data of parameters such as voltage, current, temperature, and internal resistance. This multi-scale temporal convolutional network comprises three parallel one-dimensional convolution branches, each with a convolution kernel size of 3, 5, and 7, corresponding to feature capture capabilities at different time scales. Each convolution branch consists of two one-dimensional convolutional layers, with the first layer having 32 output channels and the second layer having 64 output channels. Both layers utilize ReLU activation functions and batch normalization layers.

[0082] In some embodiments, the input of the multi-scale temporal convolutional network is a tensor of shape [B, C, T], where B is the batch size, C is the number of feature channels (including physical parameters such as voltage, current, temperature, and internal resistance), and T is the length of the time series. The multi-scale temporal convolutional network uses a dilated convolution technique to expand the receptive field to capture long-term dependencies. The dilated convolution technique has dilated ratios set to 1, 2, and 4 to ensure that the receptive field covers the entire charge and discharge cycle.

[0083] Furthermore, the outputs of each convolutional branch are weighted and fused through a channel-wise attention mechanism to generate temporal features. This attention mechanism uses learnable weight parameters to adaptively adjust the importance of features at different time scales, enhancing the model's ability to perceive key temporal patterns. The final output temporal feature dimension is [B, 64, T], preserving the original temporal resolution.

[0084] Step 22: Analyze the spatial correlation characteristics of the collected multi-source data and construct an interaction network between various physical parameters.

[0085] In some embodiments, spatial correlation analysis utilizes a graph neural network architecture, treating each physical parameter as a node of the graph neural network and the interactions between them as edges. The adjacency matrix A of the graph neural network is initialized to a fully connected form and then optimized through data-driven learning.

[0086] Specifically, the graph neural network adopts a graph convolutional network structure, which includes two layers of graph convolution layers. The mathematical expression of the first layer of graph convolution is:

[0087]

[0088] Among them, H (1) is the output of the first layer of graph convolution, X is the input feature matrix with a dimension of [N, F_in], N is the number of nodes, that is, the number of physical parameters, and F_in is the input feature dimension of each node; A is the adjacency matrix, which represents the connection relationship between parameters; D is the degree matrix, Dxy = ∑ j Axy, x, y are the node numbers, Axy represents the value of the xth row and yth column in the adjacency matrix; W (0) is a learnable weight matrix with a dimension of [F_in,64]; σ is the ReLU activation function.

[0089] The expression of the second layer of graph convolution is:

[0090]

[0091] Among them, W (1) is the weight matrix of the second layer, with a dimension of [64,32]. The final output node represents H (2)The dimension is [N,32], and the node representation can be used as an interaction network, which contains the characteristic representation of the interaction between various physical parameters.

[0092] Step 23: Fuse the temporal dimension features and spatial correlation features to construct a spatiotemporal joint representation model;

[0093] Specifically, the spatiotemporal joint representation model adopts a tensor decomposition architecture to organize time dimension features and spatial correlation features into a third-order tensor, and extracts core correlation patterns through Tucker decomposition of the third-order tensor.

[0094] First, the time dimension feature matrix (dimensions [B, 64, T]) and the spatial correlation feature matrix (dimensions [N, 32]) are combined to construct the initial space-time tensor X, with dimensions [N, T, F], where F is the combined feature dimension. The element Xijk in tensor X represents the kth feature component of the i-th physical parameter at the j-th time point.

[0095] Furthermore, Tucker decomposition decomposes the tensor X into a core tensor G and three factor matrices U (1) 、U (2) 、U (3) :

[0096] Thus, X≈G×1U (1) ×2U (2) ×3U (3) ;

[0097] Among them, G is the core tensor with dimensions [R1, R2, R3], R1, R2, R3 are the ranks of parameter dimension N, time dimension T and combined feature dimension F respectively; U (1) 、U (2) 、U (3) are factor matrices of dimensions [N, R1], [T, R2], and [F, R3], respectively; X1, X2, and X3 represent the tensor products along the parameter dimension N, the time dimension T, and the combined feature dimension F, respectively;

[0098] The decomposition process is solved by the alternating least squares method, and the optimization goal of the least squares method is to minimize the reconstruction error: min‖XG×1U (1) ×2U (2) ×3U (3) ‖ 2 ;

[0099] It can be understood that the core tensor G captures the key interaction patterns of the three dimensions of parameters, time and features, thereby forming a spatiotemporal joint representation model whose dimensions are set to [8, 10, 6], achieving effective dimensionality reduction while retaining the main information.

[0100] Step 24: Based on the joint spatiotemporal representation, a multi-level spatiotemporal correlation matrix is ​​constructed to characterize the parameter coupling relationship at different scales;

[0101] Specifically, a multi-level spatiotemporal correlation matrix is ​​derived from the result of Tucker decomposition through projection transformation. The multi-level spatiotemporal correlation matrix includes three levels of correlation matrices, representing the parameter coupling relationship at the micro, meso, and macro scales, respectively.

[0102] The calculation formula of the micro correlation matrix M_micro is:

[0103]

[0104] where G1 is the average value along the first slice of the core tensor G. M_micro has dimension [N,N] and represents the direct coupling strength between physical parameters, which is used to capture instantaneous correlations on the millisecond to second scale.

[0105] The calculation formula of the meso-correlation matrix M_meso is:

[0106]

[0107] Here, G2 is the compressed representation of the core tensor G along the third dimension. M_meso has dimensions [N, T] and represents the temporal evolution of physical parameters, which is used to capture medium-term correlations at the minute to hour level.

[0108] The calculation formula of the macro correlation matrix M_macro is:

[0109]

[0110] Among them, G3 is the global feature representation of the core tensor G. M_macro dimension is [N, F], which represents the mapping relationship between physical parameters and combined features, and is used to capture long-term correlations at the day to month level.

[0111] Step 25: weightedly fusing the multi-level correlation matrix based on preset weights to generate a final spatiotemporal correlation matrix to characterize the coupling effect between the driving condition and the charging condition of the vehicle battery;

[0112] Specifically, the final spatiotemporal correlation matrix S is obtained by weighted fusion of multi-level correlation matrices:

[0113] S=α×M_micro+β×M_meso+γ×M_macro

[0114] Among them, α, β, and γ are weight coefficients. In the actual implementation of the present invention, they are determined by cross-validation based on historical capacity decay data and are set to 0.3, 0.5, and 0.2, respectively, which can reflect the dominant influence of mesoscale correlation on battery health status.

[0115] Specifically, the spatiotemporal correlation matrix S has a dimension of [N, N+T+F] and consists of three sub-blocks: a parameter-parameter correlation block, a parameter-time correlation block, and a parameter-feature correlation block. The parameter-parameter correlation block represents the strength of the mutual influence between different physical parameters; the parameter-time correlation block represents the temporal variation of parameters; and the parameter-feature correlation block represents the mapping relationship between parameters and extracted features.

[0116] The element S_nm in the matrix S represents the correlation strength between the nth parameter and the mth dimension (which may be another parameter, time point or feature). The value range is [-1,1]. Positive values ​​indicate positive correlation, negative values ​​indicate negative correlation, and the absolute value indicates the correlation strength.

[0117] Specifically, the coupling effects between driving and charging conditions in the spatiotemporal correlation matrix S are extracted through pattern recognition in the parameter-time correlation block. The system segments the time series into driving and charging segments and analyzes the mutual influence of parameter changes under these two conditions. For example, high-intensity driving followed by rapid charging creates a specific correlation pattern in the matrix, manifesting as a strong coupling between temperature parameters and charging current.

[0118] The final generated spatiotemporal correlation matrix is ​​output in the form of structured data, which contains the following information: inter-parameter correlation strength matrix, parameter-time correlation matrix, parameter-feature correlation matrix;

[0119] Furthermore, extracting key features using an attention mechanism based on the spatiotemporal correlation matrix includes the following steps:

[0120] Step 31: Construct a multi-head self-attention mechanism, and use the parallel attention calculation units in the multi-head self-attention mechanism to enhance the spatiotemporal correlation matrix and output a feature matrix.

[0121] Specifically, firstly, the spatiotemporal correlation matrix S is linearly projected onto the query matrix Q, the key matrix K, and the value matrix V. The calculation formula is:

[0122] Q=S·WQ; K=S·WK; V=S·WV;

[0123] Among them, WQ, WK, and WV are learnable weight matrices with dimensions of [N+T+F, d]. d is the hidden dimension of the attention mechanism, which is set to 64 in this embodiment.

[0124] In this embodiment, the multi-head self-attention mechanism includes 8 parallel attention heads, each of which independently calculates attention weights and weighted features. The calculation formula for the hth attention head is:

[0125] head_h=Attention(Q·WQ_h,K·WK_h,V·WV_h);

[0126]

[0127] Among them, WQ_h, WK_h, WV_h are the projection matrices of the h-th attention head, h is a positive integer from 1 to 8, and the dimension is The softmax function normalizes the attention scores into a probability distribution; is the scaling factor to prevent the gradient from disappearing.

[0128] The output of the multi-head attention is obtained by concatenating the results of each attention head and performing a linear transformation:

[0129] MultiHead(Q,K,V)=Concat(head_1,head_2,...,head_8)·WO;

[0130] Where WO is the output projection matrix with a dimension of [d, d]. The output dimension of the multi-head self-attention mechanism is [N, d], which retains the original number of parameters but enhances the feature representation capability.

[0131] The feature matrix output by the MultiHead(Q, K, V) function is marked as F_multi. Obviously, F_multi is the spatiotemporal correlation matrix after feature enhancement.

[0132] Step 32: Based on the feature matrix output by the multi-head self-attention mechanism, a parameter-level attention mechanism is constructed to adaptively evaluate the importance of different physical parameters and identify key parameters that have a significant impact on the battery health status.

[0133] Specifically, the feature matrix F_multi output by the multi-head self-attention mechanism is first passed through a fully connected layer to generate a parameter importance score: α = sigmoid(F_multi·Wα+bα);

[0134] Among them, Wα is the weight matrix with dimension [d, N]; bα is the bias vector with dimension [N]; the sigmoid function maps the calculation results to the (0, 1) interval to represent the importance score of each physical parameter.

[0135] The output F_param of the parameter-level attention mechanism weights the features by the importance score: F_param = α⊙F_multi; where ⊙ represents element-wise multiplication, α has the dimension [N, 1], and is multiplied with F_multi through the broadcast mechanism.

[0136] Furthermore, the parameter-level attention mechanism also includes residual connections and layer normalization to stabilize the training process, for example, including: F_param_norm=LayerNorm(F_param+F_multi);

[0137] LayerNorm is a layer normalization operation that normalizes features to improve training stability. The output dimension of the parameter-level attention module is [N, d], highlighting the key parameters that significantly affect the battery health status.

[0138] Step 33: Based on the key parameters of the output of the parameter-level attention mechanism, an architecture combining multi-scale temporal convolution and attention is adopted to construct a time-scale attention mechanism to capture battery aging characteristics at different time scales.

[0139] Specifically, first, reshape the output F_param_norm of the parameter-level attention mechanism into a time series representation: F_time = Reshape(F_param_norm, [N, T, d / T]);

[0140] Among them, the Reshape operation reorganizes the features into a time series form with a dimension of

[0141] The temporal attention mechanism consists of three parallel temporal convolution branches with kernel sizes of 3, 7, and 15, corresponding to the short-term scale F_short, the medium-term scale F_medium, and the long-term scale F_long, respectively. The expressions are as follows:

[0142] F_short=Conv1 D(F_time,kernel_size=3,padding='same');

[0143] F_medium=Conv1 D(F_time,kernel_size=7,padding='same');

[0144] F_long=Conv1 D(F_time,kernel_size=15,padding='same');

[0145] The time scale features are weighted and fused through the self-attention mechanism. The specific calculation formula is as follows:

[0146] β_short=softmax(F_short·Wβ_short);

[0147] β_medium=softmax(F_medium·Wβ_medium);

[0148] β_long=softmax(F_long·Wβ_long);

[0149] F_time_att=β_short⊙F_short+β_medium⊙F_medium+β_long⊙F_long;

[0150] Among them, Wβ_short, Wβ_medium, Wβ_long are learnable weight matrices; β_short, β_medium, β_long are the attention weights of each time scale. The dimension of the output F_time_att of the time scale attention mechanism is It captures the temporal dynamics of the battery aging process.

[0151] Step 34: Based on the output of the time-scale attention mechanism, design a feature interaction attention mechanism to model the nonlinear interaction relationship between various physical parameters.

[0152] Specifically, the feature interaction attention mechanism uses a graph attention network structure to model the nonlinear interaction relationship between physical parameters. First, the output of the time scale attention mechanism F_time_att is reshaped and projected into the feature space. The specific calculation formula is as follows:

[0153] F_feat=Linear(Reshape(F_time_att,[N,d]));

[0154] Among them, Linear is a linear transformation layer that maps features to d-dimensional space.

[0155] The feature interaction attention mechanism regards each physical parameter as a node in the graph and calculates the attention coefficient between nodes through the graph attention mechanism. The specific calculation formula is as follows:

[0156] exy=LeakyReLU(a T ·[W·F_feat_x||W·F_feat_y]);

[0157]

[0158] Where W is the weight matrix with dimension [d, d]; a is the attention vector with dimension [2d]; || represents the vector concatenation operation; LeakyReLU is the activation function; γxy represents the attention coefficient of node x to node y; Nx represents the neighbor set of node x.

[0159] Then the features of each node x are updated by attention-weighted aggregation:

[0160] F_feat_new_x=σ(∑ y∈Nx γ×y·W·F_feat_y);

[0161] Here, σ is the ELU activation function. The feature interaction attention mechanism consists of two graph attention layers: the first layer contains eight attention heads, and the second layer is a single-head attention layer. The final output feature F_feat_new has a dimension of [N, d], capturing the complex nonlinear interactions between parameters.

[0162] Step 35: Fusion the output of the parameter-level attention mechanism, the output of the time-scale attention mechanism, and the output of the feature interaction attention mechanism to generate key features that characterize the nonlinear characteristics of the battery aging trajectory.

[0163] Specifically, the fusion of the key features adopts a gated feature fusion mechanism, which adaptively integrates the duration of the parameter-level attention mechanism, the time-scale attention mechanism, and the feature interaction attention mechanism. First, concatenate F_param_norm, the reshaped F_time_att, and F_feat_new, as follows:

[0164] F_concat=Concat([F_param_norm,Reshape(F_time_att,[N,d]),F_feat_new])

[0165] Among them, Concat represents the feature concatenation operation, and the feature dimension after concatenation is [N, 3d].

[0166] The gated feature fusion mechanism controls the contribution of each feature through a learnable gated unit. The formula is as follows:

[0167] g_param=sigmoid(F_concat·W_g_param+b_g_param);

[0168] g_time=sigmoid(F_concat·W_g_time+b_g_time);

[0169] g_feat=sigmoid(F_concat·W_g_feat+b_g_feat);

[0170] F_gated=g_param⊙F_param_norm+g_time⊙Reshape(F_time_att,[N,d])+g_feat⊙F_feat_new;

[0171] Among them, W_g_param, W_g_time, and W_g_feat are gating weight matrices; b_g_param, b_g_time, and b_g_feat are bias vectors; g_param, g_time, and g_feat are gating coefficients of each feature, which control the contribution ratio of different features.

[0172] The final key features are obtained through nonlinear transformation and dimensionality reduction:

[0173] F_key=Dropout(ReLU(F_gated·W_1+b_1))·W_2+b_2;

[0174] Where W_1 and b_1 are the weights and biases of the first layer transformation; W_2 and b_2 are the weights and biases of the second layer transformation; Dropout is a random dropout layer to prevent overfitting; ReLU is the activation function. The key feature dimension of the final output is [N, k], where k is the key feature dimension, set to 32.

[0175] The key feature F_key encompasses the nonlinear characteristics of the battery aging trajectory, specifically manifested in the complex interactions between parameters, aging patterns at different time scales, the impact of key parameters on health status, and abnormal patterns during charge and discharge cycles. Together, these features form a multidimensional feature space that characterizes the nonlinear characteristics of the battery aging trajectory.

[0176] Furthermore, establishing a capacity decay prediction model based on the key features and historical capacity decay data includes the following steps:

[0177] Step 41: Based on the extracted key features and historical capacity decay data, a multi-branch fusion network architecture is constructed; the multi-branch fusion network architecture includes three main components: a feature processing branch, a historical capacity branch, and a fusion prediction branch.

[0178] Step 42: In the feature processing branch, a fully connected layer is constructed to perform feature transformation on the key features and output the changed key features;

[0179] Specifically, the feature processing branch is responsible for processing key features, the historical capacity branch is responsible for processing historical capacity attenuation data, and the fusion prediction branch fuses the outputs of the two branches and generates a final prediction result.

[0180] In the feature processing branch, the key feature F_key is first input into the fully connected layer for feature transformation: F_trans = ReLU(F_key·W_fc1+b_fc1); where W_fc1 is the weight matrix with a dimension of [32, 64], b_fc1 is the bias vector with a dimension of

[64] , and ReLU is the activation function used to introduce nonlinear transformation capabilities. The output dimension after feature transformation is [N, 64], where N is the number of physical parameters.

[0181] Step 43: In the historical capacity branch, a recurrent neural network structure is constructed to capture the temporal dynamic characteristics of the capacity decay of the historical capacity decay data; based on the temporal dynamic characteristics, a temporal feature aggregation mechanism is designed to extract key information from the historical capacity decay data.

[0182] The historical capacity branch uses a Long Short-Term Memory (LSTM) network structure to process the time series characteristics of historical capacity decay data. First, the historical capacity decay data is organized into a time series format: C_hist = [C_1, C_2, ..., C_T], where C_t represents the capacity value at the tth cycle and T is the length of the historical data. The input data to the LSTM network is a combination of the capacity sequence and the corresponding number of cycles: X_t = [C_t, N_t], where N_t is the corresponding number of cycles. The forward computation of the LSTM network is: h_t, c_t = LSTM(X_t, h_{t-1}, c_{t-1}), where h_t is the hidden state at time t and c_t is the cell state at time t. LSTM is a Long Short-Term Memory network unit, and the hidden layer dimension is set to 128.

[0183] The temporal feature aggregation mechanism obtains the overall feature representation of historical capacity data by performing an average pooling operation on the hidden state sequence output by the LSTM. First, the hidden state of the LSTM network at each time step is collected: H = [h_1, h_2, ..., h_T]; where h_t is the hidden state at time t, with a dimension of

[128] . Then, the aggregated feature is obtained through an average pooling operation: F_cap = (1 / T)·Σ_t h_t; where F_cap is the aggregated historical capacity feature, with a dimension of

[128] . This mechanism can effectively capture the overall trend information during the capacity decay process and provide a reliable feature representation for subsequent predictions.

[0184] Step 44: Through the fusion prediction branch, the output of the feature processing branch and the output of the history capacity branch are spliced ​​and fused to obtain a fusion feature;

[0185] Specifically, the fusion prediction branch uses feature concatenation to integrate key features and historical capacity features. First, the output of the feature processing branch, F_trans, and the output of the historical capacity branch, F_cap, are concatenated: F_concat = Concat(F_trans, F_cap); where Concat is the feature concatenation operation and the dimension of F_concat is [N+128]. Subsequently, feature fusion is performed through a fully connected layer: F_fused = ReLU(F_concat·W_fc2+b_fc2); where W_fc2 is the weight matrix with a dimension of [N+128, 256], b_fc2 is the bias vector with a dimension of

[256] , and F_fused is the fused feature representation with a dimension of

[256] .

[0186] Step 45: Based on the concatenated and fused features, a fully connected network is constructed to predict the capacity value of the vehicle battery and calculate the health status index of the vehicle battery.

[0187] Specifically, the fully connected network contains two fully connected layers, which are used to map the fused features to the capacity prediction value. The first fully connected layer of the fully connected network is used to perform feature transformation: H_1 = ReLU(F_fused·W_1+b_1); where W_1 is a weight matrix with a dimension of [256,128], and b_1 is a bias vector with a dimension of

[128] . The second fully connected layer of the fully connected network is used to generate the capacity prediction value: C_pred = H_1·W_2+b_2; where W_2 is a weight matrix with a dimension of [128,1], b_2 is a bias vector with a dimension of [1], and C_pred is the capacity prediction value of the current vehicle battery.

[0188] Step 46: Based on the multi-branch fusion network architecture, a capacity decay prediction model is obtained by training the key features and the historical capacity decay data.

[0189] Furthermore, the method of calculating the health status index of the current vehicle battery by using the capacity decay prediction model includes:

[0190] Inputting key characteristics and historical capacity decay data of the current vehicle battery into the capacity decay prediction model to obtain a predicted capacity value of the current vehicle battery;

[0191] Based on the predicted capacity value of the current vehicle battery, a state of health (SOH) indicator of the vehicle battery is calculated using a predefined health indicator calculation formula.

[0192] In an embodiment of the present invention, SOH=(C_pred / C_rated)×100%, wherein C_pred is the predicted capacity value, C_rated is the rated capacity of the vehicle battery, and SOH represents the health of the battery in percentage.

[0193] In a further embodiment of the present invention, the health status assessment criteria are designed based on the SOH value and include the following levels: excellent (SOH ≥ 90%), good (80% ≤ SOH < 90%), average (70% ≤ SOH < 80%), poor (60% ≤ SOH < 70%), and dangerous (SOH < 60%).

[0194] Furthermore, the method for determining whether the change rate of the health status indicator exceeds a preset fluctuation range is:

[0195] If the difference between two consecutive calculation results of the health status indicator of the vehicle battery is greater than a preset difference threshold, it is determined that the change rate of the health status indicator exceeds a preset fluctuation range;

[0196] It is understandable that when the health status indicators of automotive batteries fluctuate significantly, it indicates that the battery status may have changed significantly. At this time, the model parameters need to be updated in time to capture such changes.

[0197] Therefore, the use of an incremental learning algorithm to process newly collected data in real time and update the model parameters of the capacity decay prediction model includes the following steps:

[0198] Step 51: Extract features from the newly collected multi-source data to generate feature vectors for incremental learning.

[0199] Specifically, the multi-source data is first divided into time windows, with each window containing 30 minutes of data. The same feature extraction method as in step 2 is applied to the data within each window to extract time series features and inter-parameter correlation features, generating feature vectors. These feature vectors have the same dimensions and structure as those generated from the original training data, and they reflect the performance characteristics of the battery under its current operating state.

[0200] Step 52: Construct a buffer to store the feature vectors generated within a preset time and their corresponding actual capacity values.

[0201] Specifically, the vehicle backend maintains a fixed-size buffer to store the feature vectors of the most recent 50 time windows and their corresponding actual capacity values. This buffer uses a first-in, first-out principle; when a new feature vector arrives, the oldest feature vector is removed. The feature vectors and capacity values ​​in the buffer constitute the training dataset for incremental learning, which is used to update the parameters of the capacity decay prediction model. By maintaining the buffer, the system can focus on the battery's most recent operating status and performance changes, improving the model's adaptability to the current state.

[0202] Step 53: Based on the feature vectors and actual capacity values ​​stored in the buffer, the parameters of the capacity decay prediction model are updated using the gradient descent method.

[0203] Specifically, the model parameters are updated using the feature vectors in the buffer and the actual capacity value through a mini-batch gradient descent algorithm. During the update process, the feature vector is input into the model, the mean square error between the predicted capacity value and the actual capacity value is calculated, and the gradient is calculated through the back propagation algorithm. The parameter update formula of the capacity decay prediction model is:

[0204] Among them, θ_new is the updated parameter of the capacity decay prediction model, θ_old is the original parameter of the capacity decay prediction model, and η is the learning rate, which is generally set to 0.01. is the gradient of the loss function with respect to the parameters. To prevent overfitting, L2 regularization is applied during the update process to constrain the magnitude of parameter changes. After the update is complete, the performance of the new model is evaluated using the validation set to ensure that prediction accuracy is not reduced.

[0205] In a further preferred embodiment of the present invention, a forgetting mechanism may be implemented for the buffer to adjust the weight of historical data.

[0206] Specifically, the system assigns time weights to the feature vectors in the buffer, with newer feature vectors receiving higher weights and older feature vectors receiving lower weights. The weight calculation uses a linear decay function: Where u is the position index of the feature vector in the buffer (starting from 0), and U is the buffer size. When calculating the loss function, the loss value of each sample is multiplied by the corresponding time weight to form a weighted loss function. The calculation formula is as follows:

[0207] L_weighted=Σw(u)·(y_u-f(x_u)) 2 ;

[0208] Where y_u is the actual capacity value, and f(x_u) is the capacity value predicted by the model. By using a weighted loss function, while retaining historical knowledge, it focuses more on the battery status reflected by the latest data, enhancing the model's sensitivity to the battery's current state.

[0209] Example 2

[0210] like Figure 2 As shown, a vehicle battery health status detection system based on multi-source data analysis includes a multi-source data collection module, a spatiotemporal matrix construction module, a key feature extraction module, a monitoring indicator generation module, and a model update module;

[0211] The multi-source data collection module collects multi-source data and historical capacity decay data of vehicle batteries, sends the multi-source data to the spatiotemporal matrix construction module, and sends the historical capacity decay data to the monitoring indicator generation module;

[0212] A spatiotemporal matrix construction module performs spatiotemporal correlation modeling on the multi-source data, generates a spatiotemporal correlation matrix, and sends the spatiotemporal correlation matrix to a key feature extraction module;

[0213] A key feature extraction module, which extracts key features based on the spatiotemporal correlation matrix using an attention mechanism and sends the key features to a monitoring indicator generation module;

[0214] A monitoring indicator generation module, which establishes a capacity decay prediction model based on the key characteristics and historical capacity decay data, calculates the health status indicator of the current vehicle battery, and sends the health status indicator to the model update module;

[0215] The model update module determines whether the rate of change of the health status indicator exceeds a preset fluctuation range. If so, an incremental learning algorithm is used to process the newly collected multi-source data in real time and update the model parameters of the capacity decay prediction model. Otherwise, the health status indicator is output.

[0216] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for detecting the health status of a vehicle battery based on multi-source data analysis, characterized in that: The following steps are involved: Step 1: Collect multi-source data and historical capacity decay data of vehicle batteries; Step 2: Perform spatiotemporal correlation modeling on the multi-source data to generate a spatiotemporal correlation matrix; Step 3: Based on the spatiotemporal correlation matrix, use the attention mechanism to extract key features; Step 4: Based on the key features and the historical capacity decay data, a capacity decay prediction model is established to calculate the health status index of the current vehicle battery, and determine whether the change rate of the health status index exceeds a preset fluctuation range. If so, proceed to step 5; otherwise, output the health status index; Step 5: Use an incremental learning algorithm to process the newly collected multi-source data in real time and update the model parameters of the capacity decay prediction model, and then return to step 4.

2. The method for detecting the health status of a vehicle battery based on multi-source data analysis according to claim 1, characterized in that: The collection of multi-source data includes: In an automotive battery testing environment, a multi-source sensor array is used to form a sensor network for the automotive battery. The sensor network collects the parameter values ​​of various physical parameters of the automotive battery in real time, and the parameter values ​​of the various physical parameters are combined into multi-source data; the physical parameters include voltage, current, temperature, internal resistance, and number of charge and discharge cycles.

3. The method for detecting the health status of a vehicle battery based on multi-source data analysis according to claim 2, characterized in that: The collection of historical capacity decay data includes: In the vehicle battery test environment, while collecting the multi-source data, performing a capacity test on the vehicle battery; During each capacity test, the actual available capacity value of the complete charge and discharge process is recorded, and the corresponding relationship between the current number of times and the actual available capacity value is recorded.

4. The method for detecting the health status of a vehicle battery based on multi-source data analysis according to claim 3, characterized in that: The performing of spatiotemporal correlation modeling on the multi-source data to generate a spatiotemporal correlation matrix comprises the following steps: Step 21: extracting time series features from the multi-source data to construct time dimension features; Step 22: performing spatial correlation feature analysis on the multi-source data to construct an interaction network between various physical parameters in the multi-source data; The spatial association analysis adopts a graph neural network architecture, treating each physical parameter in the multi-source data as a node of the graph neural network, and the mutual influence relationship between the physical parameters as an edge. The adjacency matrix of the graph neural network is initialized to a fully connected form, and the interaction network is obtained through data-driven learning and optimization; Step 23: Fusing the time dimension features and the spatial correlation features to construct a spatiotemporal joint representation model; The spatiotemporal joint representation model adopts a tensor decomposition architecture to organize the time dimension features and the spatial correlation features into a third-order tensor, and extracts the core correlation pattern by Tucker decomposition of the third-order tensor; Step 24: Based on the spatiotemporal joint representation, construct a multi-level spatiotemporal correlation matrix to characterize the parameter coupling relationship at different scales; the multi-level spatiotemporal correlation matrix is ​​derived from the result of Tucker decomposition through projection transformation; the multi-level spatiotemporal correlation matrix includes three levels of correlation matrices, representing the parameter coupling relationship at the micro, meso, and macro scales, respectively; Step 25: The multi-level correlation matrix is ​​weightedly fused based on preset weights to generate a spatiotemporal correlation matrix to characterize the coupling effect between the driving condition and the charging condition of the vehicle battery.

5. The method for detecting the health status of a vehicle battery based on multi-source data analysis according to claim 4, characterized in that: The method of extracting key features using an attention mechanism based on the spatiotemporal correlation matrix includes the following steps: Step 31: Construct a multi-head self-attention mechanism, and perform feature enhancement on the spatiotemporal correlation matrix and output a feature matrix through the parallel attention calculation units in the multi-head self-attention mechanism; Step 32: Based on the feature matrix output by the multi-head self-attention mechanism, construct a parameter-level attention mechanism to adaptively evaluate the importance of different physical parameters and identify key parameters that significantly affect the battery health status of the vehicle battery; Step 33: Based on the key parameters output by the parameter-level attention mechanism, a time-scale attention mechanism is constructed by combining multi-scale temporal convolution with attention to capture battery aging characteristics at different time scales. Step 34: Based on the battery aging features output by the time-scale attention mechanism, a feature interaction attention mechanism is designed to model the nonlinear interaction relationship between various physical parameters; Step 35: Fuse the output of the parameter-level attention mechanism, the output of the time-scale attention mechanism, and the output of the feature interaction attention mechanism to generate key features that characterize the nonlinear characteristics of the battery aging trajectory; the fusion of the key features adopts a gated feature fusion mechanism to adaptively integrate the durations of the parameter-level attention mechanism, the time-scale attention mechanism, and the feature interaction attention mechanism.

6. The method for detecting the health status of a vehicle battery based on multi-source data analysis according to claim 5, characterized in that: The establishing of a capacity decay prediction model based on the key features and the historical capacity decay data comprises the following steps: Step 41: Construct a multi-branch fusion network architecture; the multi-branch fusion network architecture includes three main components: a feature processing branch, a historical capacity branch, and a fusion prediction branch; Step 42: In the feature processing branch, a fully connected layer is constructed to perform feature transformation on the key features and output the changed key features; Step 43: In the historical capacity branch, a recurrent neural network structure is constructed to capture the temporal dynamic characteristics of the historical capacity decay data, and a temporal feature aggregation mechanism is designed to extract key information from the historical capacity decay data based on the temporal dynamic characteristics. Step 44: The output of the feature processing branch and the output of the historical capacity branch are concatenated and fused through the fusion prediction branch to obtain a fusion feature; Step 45: Based on the concatenated and fused fusion features, a fully connected network is constructed to predict the capacity value of the vehicle battery and calculate the health status index of the vehicle battery; the fully connected network includes two fully connected layers for mapping the fusion features to the capacity prediction value; the first fully connected layer of the fully connected network is used for feature transformation, and the second fully connected layer of the fully connected network is used for generating the capacity prediction value; Step 46: Based on the multi-branch fusion network architecture, a capacity decay prediction model is obtained by training the key features and the historical capacity decay data.

7. The method and system for detecting the health status of a vehicle battery based on multi-source data analysis according to claim 6, characterized in that: Calculating the health status index of the current vehicle battery using the capacity decay prediction model includes: Inputting key characteristics and historical capacity decay data of the current vehicle battery into the capacity decay prediction model to obtain a predicted capacity value of the current vehicle battery; Based on the predicted capacity value of the current vehicle battery, a health status index of the current vehicle battery is calculated using a predefined health status index calculation formula.

8. The method for detecting the health status of a vehicle battery based on multi-source data analysis according to claim 7, characterized in that: Determining whether the change rate of the health status indicator exceeds a preset fluctuation range includes: If the difference between two consecutive calculation results of the health status indicator is greater than a preset difference threshold, it is determined that the change rate of the health status indicator exceeds a preset fluctuation range.

9. The method for detecting the health status of a vehicle battery based on multi-source data analysis according to claim 8, characterized in that: The method of using an incremental learning algorithm to process newly collected multi-source data in real time and update the model parameters of the capacity decay prediction model includes the following steps: Step 51: Extract features from the newly collected multi-source data to generate feature vectors for incremental learning; Step 52: Construct a buffer to store the feature vectors generated within a preset time and their corresponding actual capacity values; Step 53: Based on the feature vectors and actual capacity values ​​stored in the buffer, the parameters of the capacity decay prediction model are updated using a gradient descent method.

10. A vehicle battery health status detection system based on multi-source data analysis, which is used to implement the vehicle battery health status detection method based on multi-source data analysis according to any one of claims 1 to 9, characterized in that: The detection system includes a multi-source data collection module, a spatiotemporal matrix construction module, a key feature extraction module, a monitoring index generation module and a model update module, wherein: The multi-source data collection module is used to collect multi-source data and historical capacity decay data of the vehicle battery, and send the multi-source data to the spatiotemporal matrix construction module, and send the historical capacity decay data to the monitoring indicator generation module; The spatiotemporal matrix construction module is used to perform spatiotemporal correlation modeling on the multi-source data, generate a spatiotemporal correlation matrix, and send the spatiotemporal correlation matrix to the key feature extraction module; The key feature extraction module is used to extract key features based on the spatiotemporal correlation matrix using an attention mechanism, and send the key features to the monitoring indicator generation module; The monitoring indicator generation module is used to establish a capacity decay prediction model based on the key characteristics and the historical capacity decay data to calculate the health status indicator of the current vehicle battery, and send the health status indicator to the model update module; The model update module is used to determine whether the rate of change of the health status indicator exceeds a preset fluctuation range. If so, an incremental learning algorithm is used to process the newly collected multi-source data in real time and update the model parameters of the capacity decay prediction model. Otherwise, the health status indicator is output.

Images