Battery life prediction system, method, device, storage medium and program product

Abstract

The invention relates to a battery life prediction system, method and device, a storage medium and a program product. A vector network analyzer is used for acquiring impedance spectrum data of a to-be-detected battery and outputting a Smith chart; the client is used for obtaining the Smith chart output by the vector network analyzer and sending the Smith chart and the first battery type selected by the user to the server; and the server processes the Smith chart to obtain the predicted residual capacity of the battery, and returns the predicted residual capacity to the client. According to the prediction model, line color features in a Smith chart and edge number, direction and position features of a polygon are extracted, the size of a convolution kernel is determined according to the edge length statistical value of the polygon, and the incidence relation between the prediction model and the residual capacity of the battery is determined through training. The Smith chart output by the vector network analyzer is used as input data, the prediction model is built in combination with the corresponding image recognition algorithm, the residual capacity data of the detected battery are obtained, the cost is low, the efficiency is high, and the tiny change of the battery capacity can be detected.

Description

Technical Field

[0001] The present invention relates to the technical field of battery life prediction, and in particular to a battery life prediction system, method, device, storage medium and program product. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Battery life refers to the remaining capacity of the battery. Existing technologies can use electrochemical impedance spectroscopy to reflect the degree of battery aging (such as SEI film thickening and lithium deposition) through impedance changes, and use the output curve (such as the Nyquist plot) to reflect the complex impedance to achieve detection or prediction.

[0004] However, errors in the impedance spectrum measurement process (such as noise interference, instrument accuracy deviation, etc.) will significantly change the geometry of the Nyquist plot, resulting in a decrease in the stability of feature extraction and instability of the features in the impedance spectrum data reflected by the Nyquist plot. As a result, the method of using the Nyquist plot to predict the remaining capacity of the battery is difficult to meet the needs of batch battery testing. Summary of the Invention

[0005] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a battery life prediction system, method, device, storage medium and program product, which uses the Smith chart output by a vector network analyzer as input data, combines the corresponding image recognition algorithm to build a prediction model, and obtains the remaining capacity data of the tested battery.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A first aspect of the present invention provides a battery life prediction system, comprising:

[0008] Vector network analyzer, used to obtain impedance spectrum data of the battery under test and output Smith chart;

[0009] The client is used to obtain the Smith chart output by the vector network analyzer and send it to the server together with the battery type selected by the user;

[0010] The server receives the Smith chart and battery type information, obtains the predicted remaining battery capacity through a pre-trained prediction model, and returns it to the client;

[0011] Among them, the prediction model extracts the line color features in the Smith chart, the number of edges, direction and position features of the polygon, and determines the size of the convolution kernel based on the statistical value of the polygon edge length. After training, it determines the correlation between it and the remaining battery capacity.

[0012] Furthermore, the obtained Smith chart forms a complex plane coordinate system with resistance as the horizontal axis and reactance as the vertical axis. The points or trajectories on the chart represent the impedance characteristics of the battery at different frequencies, reflecting the comprehensive response of electrolyte ion migration and electrode interface reaction processes.

[0013] Furthermore, during the training of the prediction model, the obtained Smith chart is preprocessed, specifically: the obtained Smith chart is scaled, center cropped, region of interest extracted, data enhanced and normalized.

[0014] Furthermore, during the training of the prediction model, multiple alternating convolutional layers, batch normalization layers, and pooling layers are used to extract features from the preprocessed Smith chart and associate them with the pre-calibrated remaining capacity data. After a set number of iterations, the training is completed.

[0015] Furthermore, the server is also provided with a training data upload interface, and the user sends pre-labeled battery capacity data, the corresponding Smith chart and the set battery type information to the server through the client.

[0016] Furthermore, the prediction model is equipped with a directed convolution kernel group to detect the horizontal, vertical and diagonal edges and vertices of the polygons in the Smith chart to form the number of edges, direction and position characteristics of the polygons. During this period, the size of the convolution kernel is determined according to the statistical value of the polygon side length, and the correlation between it and the remaining battery capacity is determined through training.

[0017] A second aspect of the present invention provides a battery life prediction method, comprising the following steps:

[0018] Obtain the impedance spectrum data of the battery under test and output it as a Smith chart;

[0019] The obtained Smith chart is combined with the pre-selected battery type and the pre-trained prediction model to obtain the predicted remaining capacity of the battery;

[0020] Among them, the prediction model extracts the line color features in the Smith chart, the number of edges, direction and position features of the polygon, and determines the size of the convolution kernel based on the statistical value of the polygon edge length. After training, it determines the correlation between it and the remaining battery capacity.

[0021] A third aspect of the present invention provides a computer program product, comprising computer-readable instructions, which, when executed on an electronic device, enable the electronic device to implement the above-mentioned battery life prediction method.

[0022] The fourth aspect of the present invention provides an electronic device comprising at least one processor and a memory connected to the processor, the memory being used to store a computer program; the processor being used to execute the computer program, so that the electronic device can implement the above-mentioned battery life prediction method.

[0023] A fifth aspect of the present invention provides a computer storage medium carrying one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the above-mentioned battery life prediction method.

[0024] Compared with the existing technology, one or more of the above technical solutions have the following beneficial effects:

[0025] 1. The process of obtaining the remaining capacity of the battery is converted from the traditional electrochemical method to the processing method of the Smith chart. Visual perception replaces high-precision hardware detection. Low-cost image acquisition equipment (such as industrial cameras) is combined with lightweight convolutional neural networks to reduce the detection cost by more than 90% compared with traditional capacity analyzers, solving the problem of excessive cost in existing technologies.

[0026] 2. In traditional methods, errors in the impedance spectrum measurement process (such as noise interference, instrument accuracy deviation, etc.) will significantly change the geometric shape of the output graph (such as the Nyquist plot), resulting in a decrease in the stability of feature extraction. However, the Smith chart, due to its mathematical properties of conformal mapping, is more robust to measurement errors. Even in the presence of a certain amount of noise, its graphic outline and key features (such as the center position, radius ratio, etc.) can remain basically unchanged. This feature makes the data set based on the Smith chart have higher morphological consistency, making it easier for the model to capture and learn hidden physical laws. Without additional complex data preprocessing for graphic distortion, the spatial feature extraction capability of CNN can be directly used to achieve efficient modeling, providing a more reliable input foundation for intelligent analysis of impedance spectra.

[0027] 3. It can adapt to the needs of relative capacity detection. By focusing on the rapid prediction of relative capacity (such as 80% capacity retention rate), it avoids redundant calculation of absolute capacity (kWh level), simplifies the detection process, and improves the direct applicability of the results in scenarios such as battery health management and production sorting.

[0028] 4. Through image preprocessing (sizing, green area extraction) and end-to-end model training, non-contact rapid detection of battery capacity is achieved. The time required for a single inference can be controlled in milliseconds, meeting the needs of real-time monitoring and large-scale batch processing.

[0029] 5. Traditional capacity analyzers are unable to meet the real-time battery testing needs under high production capacity. Through batch image acquisition and model parallel reasoning, it can achieve rapid screening of hundreds of batteries per minute, significantly improving production line efficiency. It is suitable for batch battery life prediction, such as batch quality inspection on battery production lines.

[0030] 6. In the battery secondary utilization scenario, it can quickly evaluate the relative capacity of the battery to determine its applicable scenario (such as energy storage, low-speed electric vehicles, etc.), avoiding the high cost and long time of traditional testing, for example, the screening of retired batteries for cascade utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0032] Figure 1 is a schematic diagram of a battery life prediction system provided by one or more embodiments of the present application;

[0033] Figure 2 This is a schematic diagram of obtaining the remaining capacity of a battery by processing a Smith chart according to one or more embodiments of the present application. DETAILED DESCRIPTION

[0034] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0035] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0036] Example 1:

[0037] The Smith chart and other impedance spectra, such as the Nyquist plot, are essentially different ways of expressing the same physical quantity, but they exhibit unique advantages in practical applications: errors in the impedance spectrum measurement process (such as noise interference, instrument accuracy deviation, etc.) can significantly change the geometry of the Nyquist plot, resulting in a decrease in the stability of feature extraction.

[0038] The Smith chart, thanks to its mathematical properties of conformal mapping, is more robust to measurement errors. Even in the presence of noise, its graphic outline and key features (such as center position and radius ratio) remain essentially unchanged. This characteristic gives Smith chart-based datasets greater morphological consistency, making it easier for convolutional neural networks (CNNs) to automatically capture and learn hidden physical laws. Without the need for complex data preprocessing to address graphic distortion, efficient modeling can be achieved directly using the spatial feature extraction capabilities of CNNs, providing a more reliable input foundation for intelligent impedance spectroscopy analysis.

[0039] Battery life prediction system, including:

[0040] Vector network analyzer, used to obtain impedance spectrum data of the battery under test and output Smith chart;

[0041] The client is used to obtain the Smith chart output by the vector network analyzer and send it to the server;

[0042] The server processes the Smith chart to obtain the predicted remaining battery capacity and returns it to the client.

[0043] A vector network analyzer (VNA) is an instrument used to measure the high-frequency characteristics of electronic networks. It mainly evaluates the scattering parameters of the device by analyzing the amplitude and phase relationship of the incident wave, reflected wave, and transmitted wave.

[0044] When used for impedance spectrum detection, the vector network analyzer injects a sine wave sweep signal into the battery, obtains a curve of the complex impedance (real part + imaginary part) changing with frequency through the sweep, and outputs it as a Smith chart.

[0045] The resulting Smith chart is a complex plane coordinate system with resistance on the horizontal axis and reactance on the vertical axis. The points or tracks on the chart represent the impedance characteristics of the battery at different frequencies, fully reflecting the comprehensive response of internal processes such as electrolyte ion migration and electrode interface reactions. At different stages of the battery's remaining life, the frequency response of the complex impedance will exhibit unique characteristics due to aging factors such as electrolyte performance degradation and changes in the electrode interface state. These characteristics are manifested on the Smith chart as specific trajectory shapes, distribution areas, or geometric parameters, such as the trajectory's position, curvature, and distance from the center of the circle. Through modeling of experimental data, a mapping relationship between the chart characteristics and the remaining life can be established: the impedance data of a new battery is usually concentrated in a certain area of ​​the chart. As the capacity decays, the data points gradually migrate to a specific direction or area, forming a characteristic pattern related to life. This correspondence is based on the overall distribution pattern of the impedance data in the complex plane, without relying on the detailed derivation of specific electrochemical mechanisms, and directly links to the battery's aging state.

[0046] Variations in battery remaining capacity are mapped to characteristic trajectories or regional distributions on the Smith chart through changes in the complex characteristics of the internal impedance (the frequency response of the real resistance and imaginary reactance). A vector network analyzer, combined with complex plane analysis of the Smith chart, transforms the complex electrochemical aging process into intuitive graphical features. By establishing a "chart feature-remaining life" model, a quantitative assessment of battery life is achieved, avoiding the one-sidedness of single impedance parameter analysis.

[0047] The Smith chart is essentially a normalized impedance coordinate system on the complex plane. Its core design goal is to visualize the real part (resistance R) and imaginary part (reactance X) of complex impedance to facilitate analysis of frequency response characteristics. The specific mapping steps are as follows:

[0048] First, the complex impedance Z = R + jX is normalized, and the normalized impedance is obtained based on the characteristic impedance Z_0 (usually the industrial standard value of 50\,\Omega):

[0049] z=\frac{Z}{Z_0}=\frac{R}{Z_0}+j\frac{X}{Z_0}=r+jx;

[0050] Where r = R / Z_0 is the normalized real part of resistance, and x = X / Z_0 is the normalized imaginary part of reactance.

[0051] After normalization, the center of the Smith chart corresponds to z = 1 + 0j (i.e., Z = Z_0), and the radius represents the relative size of the impedance modulus, which solves the problem of large differences in the absolute values ​​of impedance of different batteries and facilitates unified analysis.

[0052] The Smith chart's horizontal axis represents normalized resistance, r, and its vertical axis represents normalized reactance, x (capacitive reactance is negative, inductive reactance is positive). Each frequency point, represented by the normalized impedance z = r + jx, corresponds to a unique point on the chart. The impedance values ​​at all frequency points are connected to form an impedance trajectory curve, which intuitively reflects the impedance variation of the battery over a wide frequency range.

[0053] The vector network analyzer plots the normalized impedance z(f) at all frequency points onto a Smith chart, forming a continuous curve (trace) whose morphological characteristics are directly related to the battery aging state.

[0054] Vector network analyzers can generally output various formats, such as Nyquist plots and Smith charts. The coordinate system of the Nyquist plot is the original impedance (real part vs. imaginary part, non-normalized), while the Smith chart is the normalized impedance (based on Z_0). Secondly, the Nyquist plot is a continuous curve, which rotates clockwise from high frequency to low frequency and is prone to overlap. The Smith chart is a grid of equal resistance and equal reactance circles, which facilitates feature quantization. This characteristic means that during batch testing, the coordinate range of the Nyquist plot needs to be manually calibrated, and different batteries can easily exceed the display area. The Smith chart uses a unified normalization process to eliminate impedance scale differences caused by differences in battery models. In addition, the coordinate range of the Nyquist plot changes dynamically, requiring preprocessing to adjust the image scale, which is relatively unfriendly to the subsequent image recognition process. The Smith chart has a fixed center (1,0) and radius range (0-∞), which is more suitable for edge detection and contour extraction in batch testing scenarios.

[0055] The resulting Smith chart's trajectory intuitively depicts the impedance's frequency response: its geometric features (such as area, perimeter, and center of gravity) clearly map to the remaining capacity. Compared to the Nyquist chart, which focuses on a continuous curve from high to low frequencies, the Smith chart's normalized coordinate system (with a center of 50Ω and radius corresponding to the impedance modulus) facilitates feature quantification in image recognition (for example, the Euclidean distance from the center directly reflects changes in the impedance modulus).

[0056] The use of the Smith chart in combination with the corresponding image recognition algorithm has the advantage of being parameter-free. It does not need to rely on the precise extraction of battery equivalent circuit parameters (such as electrolyte resistance and double-layer capacitance). It directly models the trajectory morphology changes, avoiding parameter drift problems caused by differences in production processes between different batches of batteries.

[0057] Regardless of the battery type (lithium-ion, nickel-metal hydride, etc.) or aging stage, the trajectory image of the Smith chart is a two-dimensional plane curve. Features can be extracted through a unified image processing process (such as edge detection and contour extraction), significantly reducing the algorithm complexity of multi-type battery detection.

[0058] Combined with the forced retention of the green area (representing the effective impedance trajectory area) in data enhancement, the noise that may appear in batch detection (such as instrument display error and shooting angle deviation) can be filtered out to ensure the stability of feature extraction.

[0059] During the training of the image recognition algorithm model, different types of batteries (such as lithium iron phosphate and nickel metal hydride) are used, and their corresponding Smith chart trajectories are included in the training data. Through data augmentation (geometric transformation and color adjustment), possible image differences in batch detection (such as shooting angle and lighting conditions) are simulated.

[0060] The image recognition algorithm uses a lightweight CNN architecture and is optimized through dynamic learning rates and early stopping mechanisms, enabling the model to automatically capture common features of different battery models (such as trajectory deviation trends caused by aging). At the same time, it extracts the battery main area (green area) through the HSV color space, filters background noise, and reduces the interference of model differences on feature extraction.

[0061] In this embodiment, the client can be an APP pre-installed on the mobile phone, or it can be formed by accessing a website on the server, for example, Figure 1-Figure 2 As shown, after obtaining the Smith chart, you need to log in to the website with your mobile phone and use the phone's camera function to obtain the Smith chart output by the vector network analyzer. The app or website sends the Smith chart to the server. The program on the server first adjusts the image clarity and sizes it, then extracts the Smith chart based on pixel color (the core content of the Smith chart is displayed as a green curve, so the program extracts green pixels). It then uses a convolutional neural network to analyze it, and then returns the analysis results to the client.

[0062] The server receives the Smith chart and battery type information and obtains the predicted remaining battery capacity using a pre-trained prediction model, including the following steps:

[0063] Step S1: Obtain the Smith chart and pre-process it;

[0064] Preprocessing includes the following steps:

[0065] S101 proportional scaling: maintains the image aspect ratio and scales the short side to 256 pixels.

[0066] S102 Center Crop: Crop to the center area of ​​256×256 pixels.

[0067] S103 Region of Interest Extraction: Convert to HSV color space, filter the color threshold area corresponding to the region of interest (for example, obtain the green area in the Smith chart), and retain the battery body.

[0068] S104 data augmentation (training only): random flipping, rotation, and brightness adjustment to improve model generalization capabilities.

[0069] S105 Standardization: Convert to a standardized data format and perform normalization processing.

[0070] During training, the prediction model uses a data augmentation process to perform multi-dimensional transformations on the original image, including geometric transformations and color adjustments. Specifically, the geometric transformations include random scaling (0.6-1.4 times), translation (within ±20 pixels), and rotation (30° intervals); the color adjustments include brightness shift (±20) and contrast scaling (0.5-1.5 times). After the transformed image is converted to the HSV color space, a mask is created based on the set green HSV range. The original image and the mask are then ANDed together to ensure that the image content is within the green range, resulting in the enhanced image.

[0071] Step S2: Based on the pre-processed Smith chart and the battery type information selected by the user, the predicted remaining battery capacity is output. The specific steps are as follows:

[0072] S201 Feature Extraction: The preprocessed Smith chart image is fed into a convolutional neural network (CNN), which consists of multiple alternating convolutional, batch normalization, and pooling layers. The convolutional layer convolves the image with a preset 3×3 convolution kernel, with a stride of 1 and padding equal to the same, extracting low-level features such as edges and textures from the Smith chart. The batch normalization layer accelerates network convergence, introducing nonlinearity through the ReLU activation function. The pooling layer then downsamples the image using a 2×2 pooling kernel with a stride of 2, reducing the feature map size while retaining key features. After multiple layers of convolutional blocks, the high-level, abstract features of the Smith chart are ultimately extracted.

[0073] In this embodiment, based on the strong correlation between the number of sides, position and line features of the polygons in the Smith chart and slight changes in battery capacity, a geometric feature-sensitive module is designed in the network architecture and processing flow to accurately capture subtle structural differences.

[0074] In this embodiment, the Smith chart can accurately reflect subtle changes in battery capacity, so a directed convolution kernel is designed into the network. The principle is as follows: the Smith chart contains a large number of polygonal shapes, which are composed of straight lines and specific angles. Traditional circular convolution kernels (such as 3×3) are less efficient at extracting regular geometric shapes. A directed convolution kernel group is used to specifically detect the number of edges, direction, and position characteristics of polygons.

[0075] Therefore, a directional convolution kernel group is designed, including horizontal (1×3), vertical (3×1), diagonal (3×3 tilted kernel) and other direction-sensitive kernels to detect the edges and vertices of polygons.

[0076] It is also designed to adjust the convolution kernel size according to the polygon size. The kernel size is adaptively determined according to the polygon side length statistics to extract the edge and vertex features of the polygon.

[0077] S202 Network training and parameter optimization: Construct a CNN network structure containing a fully connected layer, and input the extracted features into the fully connected layer. During the training phase, the network is trained using a Smith chart image dataset labeled with the corresponding true values ​​of the remaining capacity. The mean square error (MSE) is defined as the loss function, and stochastic gradient descent (SGD) or its improved algorithm (such as Adam) is used as the optimizer. In each round of training, forward propagation calculates the error between the predicted value and the true value, and backpropagation updates the weight parameters in the network. Through continuous iterative training, the network parameters are adjusted to minimize the loss function, thereby optimizing the network's ability to learn the relationship between the Smith chart features and the remaining capacity.

[0078] S203 Calculation of remaining capacity: After training is completed, the Smith chart to be analyzed is input into the optimized CNN network. After calculation by the fully connected layer, the output layer uses a linear activation function to obtain a numerical result, which is the specific value of the predicted remaining capacity.

[0079] S204 Result Feedback: The server returns the calculated remaining capacity value to the client for the user to view and use.

[0080] This embodiment uses a convolutional neural network (CNN) to process the Smith chart. The overall network architecture is as follows:

[0081] The input size is 256×256 color images (with 3 channels). The core goal is to extract image features through alternating stacking of convolutional layers and pooling layers, combined with fully connected layers to perform binary classification. A small convolution kernel (3×3) is used to reduce parameters while maintaining local feature extraction capabilities. Batch normalization (BN) and the ReLU activation function are used to improve training stability and nonlinear expression. The number of downsampling is appropriately controlled to avoid premature loss of spatial information (suitable for small images). A dropout layer is introduced to prevent overfitting of the fully connected layers.

[0082] 1. The input layer is used to receive raw image data with a shape of (256, 256, 3) (H×W×C).

[0083] 2. Feature extraction stage (convolutional block stacking), which gradually extracts features from low-level to high-level through multiple convolutional blocks. Each block contains: convolution layer → batch normalization (BN) → ReLU activation → pooling layer (except the last block).

[0084] Convolutional block 1 includes:

[0085] Convolutional layer 1: Convolution kernel size: 3×3, number of channels (output): 32. Stride: 1, padding: size (to maintain the spatial size). Used to extract low-level features such as basic edges and textures.

[0086] The batch normalization (BN) layer is used to accelerate convergence and alleviate the gradient disappearance.

[0087] The ReLU activation function introduces nonlinearity and enhances the network's expressiveness.

[0088] Max pooling layer: pooling kernel size: 2×2, stride 2. The output size is 128×128×32 (H and W are halved, the number of channels remains unchanged).

[0089] Convolutional block 2 includes:

[0090] Convolutional layer 2: Convolution kernel size: 3×3, number of channels: 64, stride: 1, padding: same. Used to extract more complex mid-level features (such as shape and combined textures). Batch normalization (BN) + ReLU.

[0091] In the max pooling layer, the pooling kernel size is 2×2 and the stride is 2. The output size is 64×64×64.

[0092] Convolutional block 3 includes:

[0093] Convolutional layer 3: Convolution kernel size 3×3, number of channels 128, stride 1, padding: same. Used to extract high-level abstract features (such as object parts and overall structure).

[0094] Batch Normalization (BN) + ReLU.

[0095] Max pooling layer (optional):

[0096] If you are worried that the feature map size is too small, you can omit this pooling layer or switch to pooling with a stride of 1.

[0097] If retained: output size 32×32×128.

[0098] 3. Feature dimensionality reduction and classification stage

[0099] The Flatten layer converts the multidimensional feature map into a one-dimensional vector and inputs it into the fully connected layer. The output shape is 1×(32×32×128) (assuming the size after convolution block 3 is 32×32×128).

[0100] In the fully connected layer 1 (FC1), the number of neurons is 256 (which can be adjusted according to computing resources, and is recommended to be 1 / 10 to 1 / 2 of the feature dimension).

[0101] The dropout rate of the Dropout layer is 0.5 (to prevent overfitting).

[0102] The ReLU activation function is used to enhance nonlinear expression before classification.

[0103] In the fully connected layer 2 (output layer), the number of neurons is 1 (binary output). Activation function: Sigmoid outputs a probability value (0-1) and uses a threshold (e.g., 0.5) to determine the category.

[0104] The recognition model in the system is pre-trained with commonly used battery types (for example, lithium iron phosphate batteries). If the battery to be tested does not belong to the preset battery type (for example, lead-acid batteries), users are allowed to upload their own training data, including Smith charts output by vector network analyzers and remaining capacity data (in text format) obtained through other methods, to train the battery to be tested. After training, the newly added battery type is selected and the corresponding Smith chart (data to be tested) is uploaded to obtain the required remaining capacity data.

[0105] In this embodiment, after the user uploads the image data and annotation information, the image path and classification label are saved in the sqlite3 database. The image is stored in the corresponding path. The program will call the image through the path in the database and combine it with the label data for training.

[0106] The system proposed in this embodiment has the following advantages:

[0107] Significantly reduce detection costs: By replacing high-precision hardware detection with visual perception and utilizing low-cost image acquisition equipment (such as industrial cameras) combined with lightweight convolutional neural networks, detection costs can be reduced by more than 90% compared to traditional volume analyzers, solving the industry pain point of high costs in existing technologies.

[0108] Adapting to relative capacity testing needs: Focusing on the rapid prediction of relative capacity (such as a capacity retention rate of 80%), avoiding redundant calculations of absolute capacity (kWh level), simplifying the testing process, and improving the direct applicability of the results in scenarios such as battery health management and production sorting.

[0109] Improve detection convenience and efficiency: Through image preprocessing (sizing, green area extraction) and end-to-end model training, non-contact rapid detection of battery capacity is achieved. The time required for a single inference can be controlled in milliseconds, meeting the needs of real-time monitoring and large-scale batch processing.

[0110] Traditional capacity analyzers struggle to meet the real-time battery testing needs under high production capacity. This solution, through batch image acquisition and parallel model inference, can rapidly screen hundreds of batteries per minute, significantly improving production line efficiency. This solution is suitable for batch battery life prediction, such as batch quality inspection on battery production lines.

[0111] In vehicle-mounted scenarios, by integrating lightweight image sensors and edge computing modules, battery relative capacity data can be obtained in real time, providing a basis for battery balancing control without relying on external large-scale detection equipment. For example, the health status of electric vehicle battery packs can be monitored in real time.

[0112] In the battery secondary utilization scenario, this embodiment can be used to quickly evaluate the relative capacity of the battery to determine its applicable scenario (such as energy storage, low-speed electric vehicles, etc.), avoiding the high cost and long time of traditional testing, for example, the screening of retired batteries for cascade utilization.

[0113] Example 2:

[0114] The battery life prediction method includes the following steps:

[0115] Obtain the impedance spectrum data of the battery under test and output it as a Smith chart;

[0116] The obtained Smith chart is combined with the pre-selected battery type and the pre-trained prediction model to obtain the predicted battery remaining capacity.

[0117] Example 3:

[0118] A computer program product includes computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the above-mentioned battery life prediction method.

[0119] Example 4:

[0120] An electronic device includes at least one processor and a memory connected to the processor, the memory is used to store a computer program; the processor is used to execute the computer program, so that the electronic device can implement the above-mentioned battery life prediction method.

[0121] Embodiment 5:

[0122] A computer storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the above-mentioned battery life prediction method.

[0123] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A battery life prediction system, characterized in that: include: Vector network analyzer, used to obtain impedance spectrum data of the battery under test and output Smith chart; The client is used to obtain the Smith chart output by the vector network analyzer and send it to the server together with the battery type selected by the user; The server receives the Smith chart and battery type information, obtains the predicted remaining battery capacity through a pre-trained prediction model, and returns it to the client; Among them, the prediction model extracts the line color features in the Smith chart, the number of edges, direction and position features of the polygon, and determines the size of the convolution kernel based on the statistical value of the polygon edge length. After training, it determines the correlation between it and the remaining battery capacity.

2. The battery life prediction system according to claim 1, wherein: The resulting Smith chart forms a complex plane coordinate system with resistance as the horizontal axis and reactance as the vertical axis. The points or trajectories on the chart represent the impedance characteristics of the battery at different frequencies, reflecting the comprehensive response of electrolyte ion migration and electrode interface reaction processes.

3. The battery life prediction system according to claim 1, wherein: During the training of the prediction model, the Smith chart is preprocessed, specifically: scaling, center cropping, region of interest extraction, data augmentation and standardization.

4. The battery life prediction system according to claim 1, wherein: During the prediction model training, multiple alternating convolutional layers, batch normalization layers, and pooling layers are used to extract features from the preprocessed Smith chart and associate them with the pre-calibrated remaining capacity data. After a set number of iterations, the training is completed.

5. The battery life prediction system according to claim 1, wherein: The server also has a training data upload interface, and users can send pre-labeled battery capacity data, the corresponding Smith chart, and the set battery type information to the server through the client.

6. The battery life prediction system according to claim 1, wherein: The prediction model is equipped with a group of directional convolution kernels to detect the horizontal, vertical and diagonal edges and vertices of the polygons in the Smith chart to form the number of edges, direction and position characteristics of the polygons. During this period, the size of the convolution kernel is determined according to the statistical value of the polygon side length, and the correlation between it and the remaining battery capacity is determined through training.

7. A method for realizing battery life prediction based on the system according to any one of claims 1 to 6, characterized in that: The following steps are involved: Obtain the impedance spectrum data of the battery under test and output it as a Smith chart; The obtained Smith chart is combined with the pre-selected battery type and the pre-trained prediction model to obtain the predicted remaining capacity of the battery; Among them, the prediction model extracts the line color features in the Smith chart, the number of edges, direction and position features of the polygon, and determines the size of the convolution kernel based on the statistical value of the polygon edge length. After training, it determines the correlation between it and the remaining battery capacity.

8. A computer program product, characterized in that The method comprises computer-readable instructions, which, when executed on an electronic device, enable the electronic device to implement the steps in the battery life prediction method as claimed in claim 7.

9. An electronic device, characterized in that: The electronic device comprises at least one processor and a memory connected to the processor, the memory being used to store a computer program; the processor being used to execute the computer program, so that the electronic device can implement the steps in the battery life prediction method as claimed in claim 7.

10. A computer storage medium, characterized in that The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the steps in the battery life prediction method as claimed in claim 7.

Images