Lithium ion battery state real-time estimation method based on multivariate perception data

Abstract

The invention discloses a lithium ion battery state real-time estimation method based on multivariate perception data, and belongs to the technical field of lithium ion batteries. Comprising the following steps: selecting important sensing points in a battery; arranging a sensor in the battery; constructing a lithium ion battery experiment data set; performing three-dimensional interpolation reconstruction on the temperature field, performing three-way slicing to obtain a temperature image data set T, performing two-dimensional interpolation reconstruction on the pressure field, constructing a time axis to obtain a time sequence pressure image data set P, and reconstructing EIS, voltage and current data to obtain a one-dimensional feature vector; processing the data set and the feature vector by using 3D CNN, (2 + 1) D CNN, SFM and ARD; the method comprises the following steps of: extracting important global features by using Self-Attention and Cross-Attention; the transfer feature estimates battery state parameters using MLP. The method has the advantages that real-time estimation is facilitated, and high precision is still achieved when training set data is insufficient. Due to the accurate response of advanced perception to an internal field, the actual situation can be better fitted, and the estimation performance is improved by fusing multi-modal features.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion batteries, and in particular relates to a method for real-time estimation of lithium-ion battery status based on multivariate sensing data. Background Art

[0002] Lithium-ion batteries have the advantages of high energy density, low self-discharge rate, and long cycle life, and are widely used in various products. Unlike the stable environment of laboratory testing, the user's application environment and battery usage are complex, and there is no lack of extremely harsh conditions and battery abuse, which may cause the battery's performance to degrade or even fail out of control. Therefore, during the battery use process, it is necessary to monitor the battery's working condition in a timely manner and scientifically plan the battery's usage boundaries to extend its service life. The state parameters of lithium-ion batteries can reflect the battery's working state and performance information, and provide parameter support for realizing battery safety warning, fault diagnosis, and balanced composition. Therefore, accurate and real-time estimation of various state parameters has a strong practical significance.

[0003] Battery status parameters include battery health status (State of Health, SOH), battery state of charge (State of Charge, SOC) and battery remaining useful life (Remaining Useful Life, RUL). There are many parameter estimation methods: 1. Ampere-hour integration method: The ampere-hour integration method is based on testing, simple operation and high accuracy, but the battery must be fully charged and discharged in each cycle, which is difficult to meet under normal use conditions. At the same time, the status can only be fed back at the end of the cycle and cannot be fed back in real time; 2. Model-based method: The model-based method is difficult to achieve a balance between universality, accuracy and computational complexity. For example, the electrochemical model describes the internal reaction process of the battery, but contains 10 nonlinear coupled partial differential equations, which are complex in form and difficult to solve; the equivalent circuit model cannot accurately predict extreme working conditions when it is low-order, and can only be used for a special environment when it is high-order, and has low universality; 3. Filter-based method: The filter-based method has the disadvantage of not being able to take into account the complexity of model calculation and the richness of electrochemical mechanisms.

[0004] The data-driven method avoids the complex electrochemical process analysis and solution process, reduces the reliance on prior knowledge, has a simple principle and high accuracy, and is easy to fit the complex nonlinear relationship between battery state parameters and collected quantities. Currently, the characteristic quantities used for data driving are mostly voltage and current, on which various processing is performed to predict the battery state based on electrical performance parameters.

[0005] On the basis of electrical performance parameters, combining parameters with high information density and more internal conditions of the battery for network learning can better fit the actual situation, improve network accuracy and reduce errors. In recent years, with the rapid development of embedded sensors and flexible electronic technology, advanced sensing technology has become increasingly mature, and a variety of multi-physics field and high spatial resolution micro-flexible sensor arrays that can be integrated into the battery have emerged, providing technical possibilities for real-time acquisition of internal pressure, strain and temperature of the battery. The emergence of portable high-precision electrochemical impedance spectroscopy (EIS) acquisition equipment also enables us to obtain electrochemical performance parameters in real time to analyze the battery, and give full play to the information advantages of EIS on the internal material properties, interface phenomena and electrochemical reactions of the battery. Therefore, it is urgent to develop a real-time estimation method for the state of lithium-ion batteries that combines electrical performance parameters, electrochemical performance parameters and physical performance parameters obtained by sensors. Compared with the single use of electrical performance estimation, this method reduces the requirements for the amount of training set data, can obtain higher accuracy when the training set data is insufficient, and greatly shortens the acquisition time and cost of the data set. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention proposes a real-time estimation method for the state of a lithium-ion battery based on multivariate sensing data.

[0007] The purpose of the present invention is to provide a method for real-time estimation of lithium-ion battery status based on multivariate sensing data, and the specific steps are as follows:

[0008] Step 1: Select the sensor layout points. The points should include the positive and negative terminal areas, the battery center area and other key areas, and be selected in layers.

[0009] Step 2: Design sensors at important points and arrange them for M lithium-ion batteries of the same model. The sensors are preferably highly integrated, high-resolution flexible thin-film sensors, which are convenient for obtaining the various states of multiple areas inside the battery in an all-round and multi-dimensional manner with fewer devices, thus reducing the impact on the battery.

[0010] Step 3: Perform multiple cycles of complete charge and discharge on M lithium-ion batteries of the same model with sensors, and stop the experiment to obtain the experimental data set when the battery capacity decays to 80% of the nominal capacity. This includes the temperature field and pressure field data collected at multiple points during the complete charge and discharge process, the EIS, current, and voltage data collected dynamically, and the SOC, SOH, and RUL of the entire process.

[0011] Step 4: Perform three-dimensional interpolation reconstruction of temperature data based on spatial features to improve temperature resolution. Taking the orthogonal rectangular coordinate system as an example, the cylindrical coordinate system is the same as the spherical coordinate system. Use the X, Y, and Z coordinate axes as the reference directions to perform three-dimensional multi-layer slicing on the reconstructed network, and store the slice section temperature field map to form a temperature image dataset T, including the sub-dataset T under the three-dimensional slicing x , T y , T z The pressure data is reconstructed by two-dimensional interpolation to obtain a high-resolution pressure field map, and the time axis is constructed to obtain a video-like sampling time series pressure image dataset P. Both the temperature image and the pressure image are single-channel images, which is convenient for simplifying subsequent image processing and reducing computational overhead.

[0012] Step 5: Reconstruct the EIS data, voltage, and current at multiple frequencies to obtain a one-dimensional feature vector [real 1 ,real 2 ,……,real n ,imag 1 ,imag 2 ,……,imag n ,V 1 ,V 2 ,……,V n ,I 1 ,I 2 ,……,I n ], where real i is the real part of EIS at the ith frequency point, imag i is the imaginary part of the EIS at the ith frequency point, V i with I i The voltage and current when collecting the EIS data at this frequency point are arranged in time sequence to form the electrochemical and electrical performance data set E.

[0013] Step 6: Dataset division and normalization: divide the data sets T, P, E and SOC, SOH, RUL described in steps 4 and 5 into training sets and test sets, and use the Z-score normalization method for normalization.

[0014] Step 7: Process the temperature image dataset T. Take the temperature image sub-dataset T in the x direction under three-dimensional slicing x For example, take T x Indicates the current slice, T x-1 represents the previous adjacent slice in the x direction, T x+1 represents the next adjacent slice in the x direction, for T x-1 , T x , T x+1Using a 3D CNN with residual connections, a 3D convolution kernel of size i×j×3 is used to slide on the 3D image sequence, and the depth is gradually increased through convolution calculation to enrich the semantics and extract local features. The features are fused and enhanced through the Spatial Fusion Module (SFM). SFM first merges three consecutive images and converts 3×H×W×D into H×W×3D, where H is the height of the image, W is the width of the image, and D is the number of channels of the image, denoted by T. SF , using average pooling and maximum pooling to aggregate the spatial information of the feature map, expressed as:

[0015] SFM(T SF )=T SF ×σ(MLP(AvgPool(T SF ))+MLP(MaxPool(T SF )))

[0016] The spatial information is then input into a shared network consisting of a multilayer perceptron (MLP) and hidden layers.

[0017] In the formula, SFM represents the spatial fusion module, T SF represents three adjacent continuous images in a certain direction in the temperature image dataset T used for the spatial fusion module, σ is the sigmoid function, MLP represents the multi-layer perceptron, AvgPool represents the average pooling, and MaxPool represents the maximum pooling.

[0018] To T x , T y , T z All convolutional layers are processed as above and batch normalized.

[0019] Step 8: Process the pressure image dataset P. Use (2+1)D convolution with residual connection to factorize the convolution to process the spatial and temporal dimensions respectively, reducing the size of parameters and weight matrices. The convolution kernel of this convolution method consists of a size of N i-1 ×1×d×d two-dimensional convolution kernel and size M i ×t×1×1 temporal convolution kernel, t represents the time range and d represents the spatial width and height. Since it is a single-channel image, N i-1 = 1. Hyperparameter M i Determines the intermediate subspace dimension of the signal between spatial convolution and temporal convolution, and selects Where M i represents the number of output channels of the temporal convolution kernel, t represents the time range, d represents the spatial width and height, N i-1Indicates the number of output channels of the previous convolution layer, where N i-1 =1,N i Indicates the number of output channels of the current convolutional layer. Batch normalization is applied to all convolutional layers to accelerate convergence and simplify the parameter adjustment process, while global spatiotemporal average pooling is applied to the final convolution tensor.

[0020] Step 9: Extract the frequencies with high contribution rate of EIS. Add the covariance function kernel of Automatic Relevance Determination (ARD) to Gaussian Process Regression (GPR), and optimize the parameters by the maximum likelihood method. After important feature identification, remove the redundant columns from the 4n columns of real part, imaginary part, voltage and current data, and simplify the network.

[0021] Step 10: Add a self-attention mechanism to the temperature, pressure, electrochemical and electrical performance processing modules, assign weights to each feature, and extract important global features.

[0022] Step 11: Use the fully connected layer to map the global important features of temperature and pressure to a one-dimensional vector, with the number of important features as the vector length. After concatenating the two vectors, use the fully connected layer to adjust the dimension to the same as the feature vector dimension in the dataset E after step 9, forming the query vector Q.

[0023] Step 12: Use the high-contribution electrochemical and electrical performance feature vectors obtained in step 9 as the key vector K and value vector V, and use the cross-attention mechanism to establish the cross-correlation relationship between different modes. Fuse the output of the cross-attention mechanism with other features.

[0024] Step 13: Pass the fused features to the Multilayer Perceptron (MLP) to estimate the lithium-ion battery state parameters SOH, SOC, and RUL, and denormalize to obtain the final result.

[0025] The beneficial effects of the present application are as follows: the present invention is based on an implantable battery sensor and a portable electrochemical impedance spectroscopy monitoring method, combines multiple physical fields inside the battery, electrochemical impedance spectrum information and electrical performance information, integrates multiple modes, and takes advantage of the high information density of the direct state parameters inside the battery, alleviating the problem of training difficulties when the training set data is insufficient while improving the estimation accuracy. The internal physical field is spatially sliced ​​and temporally sliced, and the slice images are directly processed using 3D CNN and (2+1)D CNN methods to facilitate real-time estimation. The ARD method is used to screen important frequencies to reduce EIS acquisition time. Finally, a cross-correlation relationship is established, and the fusion of various important global features enhances the estimation performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0027] Figure 1 This is a flow chart of the method for real-time estimation of lithium-ion battery status based on multivariate sensing data of the present invention.

[0028] Figure 2 This is a model structure diagram of the real-time estimation method of lithium-ion battery status based on multivariate sensing data of the present invention.

[0029] Figure 3 It is a schematic diagram of a system of a lithium-ion battery state real-time estimation method based on multi-sensing data of the present invention.

[0030] Figure 4 It is a schematic diagram of the temperature and pressure data processing method of the real-time estimation method of the lithium-ion battery state based on multi-sensing data of the present invention.

[0031] In the figure: 1. Schematic diagram of sensor insertion position; 2. Flexible film sensor; 3. Temperature sensor; 4. Pressure sensor. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0033] It should be noted that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatus.

[0034] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0035] Combination Figure 1 , Figure 2 The method for real-time estimation of lithium-ion battery status based on multivariate sensing data of the present invention comprises the following specific steps:

[0036] Step 1: Select important points for sensor layout. The points should include the positive and negative terminal areas, the battery center area and other key areas. In order to reduce the impact on the internal structure of the battery, 2-3 planes are preferred, and important points are selected on the planes.

[0037] Step 2: See Figure 3 , design and arrange sensors for important points. Implantable sensors are preferably highly integrated, high-resolution, multi-physical field coupled flexible thin film sensors, which are convenient for using fewer devices, all-round, and multi-dimensional to know the status of multiple areas inside the battery. Its thinner volume and softer characteristics are convenient for conforming to the internal structure of the battery, which can reduce the impact on battery performance. The number of sensors is the same as the number of planes selected in step 1, and each sensor monitoring unit is arranged on the plane according to important points.

[0038] Step 3: Refer to Figure 3 , a multi-cycle complete charge and discharge experiment was conducted on M square soft-pack lithium-ion batteries of the same model with sensors arranged thereon, 1C high-rate discharge was selected to accelerate aging, the experimental time was compressed, and the experiment was stopped when the battery capacity decayed to 80% of the nominal capacity to obtain the experimental data set. Including the temperature field and pressure field data collected by the sensor during the complete charge and discharge process, the EIS, current, and voltage data dynamically collected by the EIS acquisition system, and the full-process SOC, SOH, and RUL collected by the battery performance acquisition system. Ensure the consistency of each collected quantity in time.

[0039] Step 4: Read Figure 4 , based on the spatial features, the temperature data is reconstructed by three-dimensional interpolation to improve the temperature resolution, and the reconstruction network is sliced ​​in three directions with the X, Y, and Z coordinate axes as the reference directions, where X is the battery width, Y is the battery length, and Z is the battery thickness. Preferably, the number of slices in each direction is not less than 5. The temperature field map of the slice section is stored to form a temperature image data set T, which includes the sub-datasets T under the three-way slices x , T y , T z Since pressure is a vector, only two-dimensional interpolation reconstruction is performed in the X and Y directions according to the pressure data modulus to obtain a high-resolution pressure field map, and a time axis is constructed to obtain a video-like sampling time series pressure image dataset P. Both the temperature image and the pressure image are single-channel images, and the color depth is used to represent the value, which is convenient for simplifying subsequent image processing and reducing computational overhead.

[0040] Step 5: Reconstruct the EIS data, voltage, and current at multiple frequencies to obtain a one-dimensional feature vector [real1, real2, …, realn, imag1, imag2, …, imagn, V1, V2, …, Vn, I1, I2, …, In], where reali is the real part of the EIS at the i-th frequency point, imagi is the imaginary part of the EIS at the i-th frequency point, and V i with I i The voltage and current when the EIS data of this frequency point is collected are arranged in time sequence to form the electrochemical and electrical performance data set E. That is, the first feature vector is collected for n seconds, the second feature vector is collected for 2n seconds, and so on.

[0041] Step 6: Dataset division and normalization: divide the data sets T, P, E and SOC, SOH, RUL described in steps 4 and 5 into training sets and test sets, taking 70% as the training set and 30% as the test set, and use the Z-score normalization method for normalization.

[0042] Step 7: Process the temperature image dataset T. Take the temperature image sub-dataset T in the x direction under three-dimensional slicing x For example, take T x Indicates the current slice, T x-1 represents the previous adjacent slice in the x direction, T x+1 represents the next adjacent slice in the x direction, for T x-1 , T x , T x+1 Use a 3D CNN with residual connections, and use a 3D convolution kernel of size i×j×3 to slide on the 3D image sequence. When selecting i and j, it is preferred to be consistent with the aspect ratio of the battery. Zero Padding is preferred to protect edge information. Through convolution calculation, the depth is gradually increased to enrich the semantics and extract local features. The features are fused and enhanced through the Spatial Fusion Module (SFM). SFM first merges three consecutive images and converts 3×H×W×D into H×W×3D, where H is the height of the image, W is the width of the image, and D is the number of channels of the image, denoted as T. SF , using average pooling and maximum pooling to aggregate the spatial information of the feature map, expressed as:

[0043] SFM(T SF )=T SF ×σ(MLP(AvgPool(T SF ))+MLP(MaxPool(T SF )))

[0044] The spatial information is then input into a shared network consisting of a multilayer perceptron (MLP) and hidden layers.

[0045] In the formula, SFM represents the spatial fusion module, T SF represents three adjacent continuous images in a certain direction in the temperature image dataset T used for the spatial fusion module, σ is the sigmoid function, MLP represents the multi-layer perceptron, AvgPool represents the average pooling, and MaxPool represents the maximum pooling.

[0046] Step 8: Process the pressure image dataset P. Use (2+1)D convolution with residual connection to factorize the convolution to process the spatial and temporal dimensions respectively, reducing the size of parameters and weight matrices. The activation function is preferably the ReLU function. The convolution kernel of this convolution method consists of a size of N i-1 ×1×d×d two-dimensional convolution kernel and size M i ×t×1×1 temporal convolution kernel, t represents the time range and d represents the spatial width and height. Since it is a single-channel image, N i-1 = 1. Hyperparameter M i Determines the intermediate subspace dimension of the signal between spatial convolution and temporal convolution, and selects Where M i represents the number of output channels of the temporal convolution kernel, t represents the time range, d represents the spatial width and height, N i-1 Indicates the number of output channels of the previous convolution layer, where N i-1 =1,N i Indicates the number of output channels of the current convolutional layer. Batch normalization is applied to all convolutional layers to accelerate convergence and simplify the parameter adjustment process, while global spatiotemporal average pooling is applied to the final convolution tensor.

[0047] Step 9: Extract the frequencies with high contribution rate of EIS. Add the covariance function kernel of Automatic Relevance Determination (ARD) to Gaussian Process Regression (GPR), and optimize the parameters by the maximum likelihood method. After important feature identification, remove the redundant columns from the 4n columns of real part, imaginary part, voltage and current data, and simplify the network. In order to maximize the estimation accuracy, retain no less than 20 columns.

[0048] Step 10: Add a self-attention mechanism to the temperature, pressure, electrochemical and electrical performance processing modules, assign attention weights to each feature, extract important global features, and eliminate irrelevant redundant features to accelerate subsequent calculations.

[0049] Step 11: Use the fully connected layer to map the global important features of temperature and pressure to a one-dimensional vector, with the number of important features as the vector length. After concatenating the two vectors, use the fully connected layer to adjust the dimension to be consistent with the feature vector dimension in the dataset E after step 9, forming the query vector Q.

[0050] Step 12: Use the high-contribution electrochemical and electrical performance feature vectors obtained in step 9 as the key vector K and value vector V, and use the cross-attention mechanism to establish the cross-correlation relationship between different modes. Fuse the output of the cross-attention mechanism with other features.

[0051] Step 13: Pass the fused features to the Multilayer Perceptron (MLP) to estimate the lithium-ion battery state parameters SOH, SOC, and RUL. The perceptron hidden layer is preferably 4 layers, and the activation function is preferably the ReLU activation function. Denormalization is performed to obtain the final result.

[0052] So far, the embodiments of the present disclosure have been described in detail in conjunction with the drawings. It should be noted that the implementation methods not shown or described in the drawings or the body of the specification are all forms known to ordinary technicians in the relevant technical field and are not described in detail.

[0053] In summary, the present disclosure provides a flexible sensor array capable of monitoring multiple points and multiple parameters inside a lithium battery. The array is obtained through the design and preparation method of flexible electronic technology, and includes a number of sensor units, each of which integrates more than two sensors. Compared with the existing battery internal monitoring technology, the flexible sensor array provided by the present disclosure can simultaneously monitor multiple parameters in multiple areas inside a lithium battery, and has a higher spatial resolution.

[0054] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for real-time estimation of lithium-ion battery status based on multivariate sensing data, characterized in that: The specific steps are as follows: S1: Select important points for sensor layout. The points should include the positive and negative terminal areas and the battery center area. Multiple layers should be selected in layers. S2: Design implantable sensors for important points and arrange implantable sensors for M lithium-ion batteries of the same model; S3: Perform multiple cycles of complete charge and discharge on M lithium-ion batteries of the same model equipped with sensors. When the battery capacity decays to 80% of the nominal capacity, stop the experiment to obtain the experimental data set. The experimental data set includes the temperature field and pressure field data collected at multiple points during the complete charge and discharge process, the EIS, current, and voltage data collected dynamically, and the SOC, SOH, and RUL of the whole process. S4: Process the temperature field data and the pressure field data to construct a three-dimensional slice image dataset T based on three-dimensional reconstruction and a video sampling time series pressure image dataset P based on two-dimensional reconstruction respectively; S5: Reconstruct the EIS, voltage, and current data at multiple frequencies and arrange them in time sequence to form an electrochemical and electrical performance data set E; S6: data set division and normalization processing, dividing the data sets T, P, E and SOC, SOH, RUL described in S4 and S5 into training sets and test sets, and performing normalization processing; S7: Process the temperature image dataset T, use a 3D CNN with residual connections for each three adjacent slices, input the results into the spatial fusion module to fuse and enhance the features, and perform batch normalization on all convolutional layers in the process; S8: Process the pressure image dataset P, use (2+1)D convolution with residual connection, factorize the convolution to process the spatial and temporal dimensions separately, apply batch normalization to all convolutional layers to accelerate convergence and simplify the parameter adjustment process, and apply global spatiotemporal average pooling to the final convolution tensor; S9: Add the covariance function kernel determined by autocorrelation to Gaussian process regression, extract the EIS high contribution frequency, remove the redundant columns from the 4n columns of real, imaginary, voltage, and current data through important feature identification, and simplify the network; S10: Add the self-attention mechanism to extract globally important features according to the attention weights; S11: Use a fully connected layer to map the global important features of temperature and pressure to a one-dimensional vector, with the number of important features as the vector length. After concatenating the two vectors, use a fully connected layer to adjust the dimension to be the same as the feature vector dimension in the dataset E after S9, forming the query vector Q; S12: Use the high contribution rate feature vector obtained in S9 as the key vector K and value vector V, use the cross-attention mechanism to establish the cross-correlation relationship between different modalities, and fuse the output of the cross-attention mechanism with other features; S13: The fused features are passed to a multi-layer perceptron to estimate the lithium-ion battery state parameters SOH, SOC, and RUL, and denormalized to obtain the final result.

2. The method for real-time estimation of lithium-ion battery status based on multivariate sensing data according to claim 1, characterized in that: The implantable sensor adopts a highly integrated, high-resolution, multi-physical field coupled flexible thin film sensor.

3. The method for real-time estimation of lithium-ion battery status based on multivariate sensing data according to claim 2 is characterized in that: The number of the planar flexible film sensors is 2-3, and each sensor monitoring unit is arranged on the plane according to important points.

4. The method for real-time estimation of lithium-ion battery status based on multivariate sensing data according to claim 1, characterized in that: Step S4 processes the temperature field data and the pressure field data. The specific steps are as follows: S401: Perform three-dimensional interpolation reconstruction on the temperature data. Taking the orthogonal rectangular coordinate system as an example, the cylindrical coordinate system and the spherical coordinate system are the same. The reconstruction network is sliced ​​in three directions and multilayers with the X, Y, and Z coordinate axes as the reference directions. The temperature field map of the slice section is stored to form a temperature image data set T, which includes the sub-datasets T under the three-way slice. x , T y , T z . S402: Perform two-dimensional interpolation reconstruction on the pressure data to obtain a high-resolution pressure field map, construct a time axis, and obtain a video-like sampling time-series pressure image dataset P.

5. The method for real-time estimation of lithium-ion battery status based on multivariate sensing data according to claim 4 is characterized in that: The temperature image and the pressure image are both single-channel images, which facilitates simplifying subsequent image processing and reducing computational overhead.

6. The method for real-time estimation of lithium-ion battery status based on multivariate sensing data according to claim 1, characterized in that: Step S5 processes the EIS data and the current and voltage data, specifically: reconstructing the EIS, voltage, and current data at multiple frequencies to obtain a one-dimensional feature vector [real1, real2, ..., real n ,imag1,imag2,……,imag n ,V1,V2,……,V n ,I1,I2,……,I n ], where real i is the real part of EIS at the ith frequency point, imag i is the imaginary part of the EIS at the ith frequency point, V i with I i The voltage and current when collecting the EIS data at this frequency point are arranged in time sequence to form the electrochemical and electrical performance data set E.

7. The method for real-time estimation of lithium-ion battery status based on multivariate sensing data according to claim 1, characterized in that: In step S6, the Z-score normalization method is used to perform normalization processing.

8. The method for real-time estimation of lithium-ion battery status based on multivariate sensing data according to claim 1, characterized in that: Step S7 processes the temperature image data set T, and the specific steps are as follows: S701: For the temperature image sub-dataset T in the x direction under three-way slicing x , T x Indicates the current slice, T x-1 represents the previous adjacent slice in the x direction, T x+1 represents the next adjacent slice in the x direction, for T x-1 , T x , T x+1 Use a 3D CNN with residual connections, use a 3D convolution kernel of size i×j×3 to slide on the 3D image sequence, perform batch normalization on all convolution layers, gradually increase the depth through convolution calculation to enrich the semantics, extract local features, and perform T y , T z All are processed as above; S702: The features are fused and enhanced through the Spatial Fusion Module (SFM). SFM first merges three consecutive images and converts 3×H×W×D into H×W×3D, where H is the height of the image, W is the width of the image, and D is the number of channels of the image, denoted by T. SF , using average pooling and maximum pooling to aggregate the spatial information of the feature map, expressed as: SFM(T SF )=T SF ×σ(MLP(AvgPool(T SF ))+MLP(MaxPool(T SF ))) The spatial information is then input into a shared network consisting of a multilayer perceptron (MLP) and hidden layers. In the formula, SFM represents the spatial fusion module, T SF represents three adjacent continuous images in a certain direction in the temperature image dataset T used for the spatial fusion module, σ is the sigmoid function, MLP represents the multi-layer perceptron, AvgPool represents the average pooling, and MaxPool represents the maximum pooling.

9. The method for real-time estimation of lithium-ion battery status based on multivariate sensing data according to claim 1, characterized in that: Step S8 processes the pressure image data set P, and the specific method is as follows: Using (2+1)D convolution with residual connection, the convolution is factorized to process the spatial and temporal dimensions respectively. The convolution kernel of this convolution method consists of size N i-1 ×1×d×d two-dimensional convolution kernel and size M i The convolution kernel is composed of ×t×1×1, where t represents the time range and d represents the spatial width and height. Since it is a single-channel image, N i-1 =1, hyperparameter M i Determines the intermediate subspace dimension of the signal between spatial convolution and temporal convolution, and selects Batch normalization is applied to all convolutional layers to accelerate convergence and simplify the parameter adjustment process, while global spatiotemporal average pooling is applied to the final convolution tensor; Where M i represents the number of output channels of the temporal convolution kernel, t represents the time range, d represents the spatial width and height, N i-1 Indicates the number of output channels of the previous convolution layer, where N i-1 =1,N i Indicates the number of output channels of the current convolutional layer.

Images