A method and device for predicting the remaining service life of a supercapacitor
By combining image fusion technology with convolutional neural networks, the lag problem in predicting the remaining lifespan of supercapacitors is solved, enabling early, fast, and high-precision lifespan prediction. This improves testing efficiency and reduces costs, and is applicable to predicting the remaining lifespan of supercapacitors.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI TECH UNIV
- Filing Date
- 2026-04-08
- Publication Date
- 2026-07-03
AI Technical Summary
Existing methods for predicting the remaining lifespan of supercapacitors rely on complete lifecycle data, which makes it difficult to make rapid and accurate predictions in the early stages and fails to meet the rapid assessment needs of practical engineering projects.
A method based on image fusion technology and early cycle data is adopted. By acquiring the time-series recording curves of supercapacitors during the early charge-discharge cycle, a two-dimensional spatial image is constructed and channels are stacked. Convolutional neural networks are used for feature extraction, and the remaining service life prediction results are output.
This enables early, rapid, and high-precision prediction of the remaining lifespan of supercapacitors, improving testing efficiency, reducing testing costs, and providing support for the screening and optimized control of energy storage systems.
Smart Images
Figure CN122336490A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of supercapacitor technology, specifically relating to a method and device for predicting the remaining service life of a supercapacitor. Background Technology
[0002] As a novel energy storage technology, supercapacitors (including double-layer capacitors, lithium-ion capacitors, sodium-ion capacitors, etc.) have advantages such as high power density, fast charging speed, long service life, and wide operating temperature range. Therefore, energy storage systems using only supercapacitors, as well as hybrid energy storage systems composed of supercapacitors and other energy storage devices, are widely used in critical infrastructure such as data centers, intelligent computing centers, electrified transportation vehicles, and renewable energy systems.
[0003] To ensure the normal operation of such facilities and improve the safety and economy of supercapacitor systems during operation, it is necessary to effectively assess the health status and remaining service life of supercapacitor systems. Remaining service life refers to the number of complete charge-discharge cycles a supercapacitor can withstand before its key performance parameters, such as capacitance or internal resistance, decay to the failure threshold. Accurately predicting remaining service life is crucial for enabling predictive maintenance of supercapacitors.
[0004] In existing technologies, methods for predicting the remaining lifespan of supercapacitors mostly rely on training and prediction using complete lifecycle data. However, supercapacitors have extremely long cycle lifespans (ranging from hundreds of thousands to millions of cycles). If predictions are made after the complete data has been accumulated, it will consume a lot of time and testing costs, making it difficult to meet the needs of early and rapid assessment in practical engineering.
[0005] Therefore, there is an urgent need for a method that can quickly and accurately predict the remaining lifespan of supercapacitors in the early cycling stage, in order to improve testing efficiency, reduce testing costs, and provide support for the screening and optimized control of energy storage systems. Summary of the Invention
[0006] In view of the shortcomings of the prior art described above, this application designs a method and device for predicting the remaining service life of supercapacitors based on image fusion technology and early cycle data. This method can quickly and accurately predict the remaining service life of supercapacitors in the early cycle stage, thereby improving testing efficiency, reducing testing costs, and providing support for the screening and optimized control of energy storage systems.
[0007] To achieve the above and other related objectives, this application provides a method for predicting the remaining service life of a supercapacitor, comprising: Obtain at least one time-series recording curve of the target supercapacitor during the early charge-discharge cycle process; A two-dimensional spatial image is constructed based on the time-series recording curve; Channel stacking is performed on the two-dimensional spatial image to generate a multi-channel fused image; The multi-channel fused image is input into a pre-trained convolutional neural network for feature extraction, and the remaining lifespan prediction result of the target supercapacitor is output.
[0008] In an optional embodiment of this application, constructing a two-dimensional spatial image based on the time-series recording curve includes: A sliding window approach is used to divide the time series data corresponding to each time series record curve into multiple subsequences; The subsequences are mapped to vectors, and the distance between any two vectors is calculated to construct a distance matrix; The distance matrix is converted into the two-dimensional spatial image.
[0009] In an optional embodiment of this application, before dividing the time-series data corresponding to each time-series record curve into multiple sub-sequences using a sliding window approach, the method further includes: The time-series recording curve is normalized in size, and the vector corresponding to the time-series recording curve is normalized to an integer multiple of one hundred data points, and the data is normalized to the range of [0,1].
[0010] In an optional embodiment of this application, mapping the subsequence to vectors and calculating the distance between any two vectors to construct a distance matrix includes: The subsequences are mapped to vectors, and the Euclidean distance between any two vectors is calculated to construct the distance matrix.
[0011] In one alternative embodiment of this application, the convolutional neural network includes AlexNet.
[0012] In one optional embodiment of this application, channel stacking is performed based on the two-dimensional spatial image to generate a multi-channel fused image, including: The two-dimensional spatial image is downsampled to scale its size to the size required by the convolutional neural network. The downsampled two-dimensional spatial image is stacked into channels to generate a multi-channel fused image.
[0013] In an optional embodiment of this application, obtaining at least one time-series recording curve of the target supercapacitor during the early charge-discharge cycle includes: Obtain three time-series recording curves of the target supercapacitor during several early charge-discharge cycles, wherein the three time-series recording curves include a complete voltage curve during the charge-discharge process, a charging capacity curve during the charging process, and a discharging capacity curve during the discharging process.
[0014] In one optional embodiment of this application, channel stacking is performed based on the two-dimensional spatial image to generate a multi-channel fused image, including: The three two-dimensional spatial images corresponding to the complete voltage curve during the charging and discharging process, the charging capacity curve during the charging process, and the discharging capacity curve during the discharging process are stacked in channels to generate a three-channel fused image.
[0015] In an optional embodiment of this application, the pre-trained convolutional neural network is trained in the following manner: Multiple training samples are acquired, each training sample including a multi-channel fused sample image and a corresponding true cycle lifetime label; The multi-channel fused sample image is used as input, and the real cycle lifetime label is used as the label to train the initial convolutional neural network model.
[0016] To achieve the above and other related objectives, this application provides a device for predicting the remaining lifespan of a supercapacitor, comprising: The data acquisition module is used to acquire at least one time-series recording curve of the target supercapacitor during the early charge-discharge cycle process; The image construction module is used to construct a two-dimensional spatial image based on the time-series recording curve; An image fusion module is used to stack channels based on the two-dimensional spatial image to generate a multi-channel fused image; The lifespan prediction module is used to input the multi-channel fused image into a pre-trained convolutional neural network for feature extraction and output the remaining lifespan prediction result of the target supercapacitor.
[0017] This application discloses a method for predicting the remaining lifespan of a supercapacitor, comprising: acquiring at least one time-series recording curve of a target supercapacitor during its early charge-discharge cycle; constructing a two-dimensional spatial image based on the time-series recording curve; stacking channels of the two-dimensional spatial image to generate a multi-channel fused image; inputting the multi-channel fused image into a pre-trained convolutional neural network for feature extraction; and outputting a prediction result of the remaining lifespan of the target supercapacitor. By extracting the time-series recording curve of the supercapacitor during its early charge-discharge cycle to predict its remaining lifespan, this method solves the problem of prediction lag caused by the long-term data accumulation required by traditional methods. By constructing a multi-channel fused image from the time-series recording curve and inputting it into a convolutional neural network, the method fully leverages the advantages of convolutional neural networks in feature extraction, deeply mining the local temporal correlation and multi-channel coupling features in the image, improving the prediction accuracy of the supercapacitor's remaining lifespan, thereby achieving early, rapid, and high-precision prediction of the supercapacitor's remaining lifespan, providing support for the selection and optimized control of energy storage systems. Attached Figure Description
[0018] Figure 1 A flowchart illustrating a method for predicting the remaining lifespan of a supercapacitor provided in this application; Figure 2 This is a three-channel fused image of a single supercapacitor cell; Figure 3 This is a schematic diagram of a supercapacitor remaining life prediction device provided in this application. Detailed Implementation
[0019] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.
[0020] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the illustrations only show the components related to this application and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.
[0021] Figure 1 This application illustrates a method for assessing the health status of a supercapacitor system based on image fusion technology and early cyclic data, according to an exemplary embodiment. The method includes: Step S10: Obtain at least one time-series recording curve of the target supercapacitor during the early charge-discharge cycle process; Step S20: Construct a two-dimensional spatial image based on the time-series recording curve; Step S30: Stack the channels according to the two-dimensional spatial image to generate a multi-channel fused image; Step S40: Input the multi-channel fused image into a pre-trained convolutional neural network for feature extraction, and output the prediction result of the remaining service life of the target supercapacitor.
[0022] The following will combine Figure 1 The technical solution of this application will be described in detail below.
[0023] First, step S10 is executed to obtain at least one time-series recording curve of the target supercapacitor during its early charge-discharge cycle. It should be noted that to address the issue that a single feature cannot comprehensively reflect the degradation performance of the supercapacitor, multiple time-series recording curves are typically selected. By extracting the time-series recording curves of the supercapacitor during its early charge-discharge cycle, the remaining lifespan of the supercapacitor can be predicted, overcoming the prediction lag caused by the long-term data accumulation required by traditional methods.
[0024] In some embodiments, the specific steps for obtaining at least one time-series recording curve of the target supercapacitor during its early charge-discharge cycles are as follows: obtaining three time-series recording curves of the target supercapacitor during several early charge-discharge cycles, wherein the three time-series recording curves include, for example, a complete voltage curve during the charge-discharge process, a charging capacity curve during the charging process, and a discharging capacity curve during the discharging process. It should be noted that in other embodiments, the time-series recording curves may also include, for example, a charge-discharge current curve.
[0025] As an example, taking the first five charge-discharge cycles of the target supercapacitor as the object of study, three time-series recording curves are extracted, namely the complete voltage V during the charge-discharge process, the charging capacity C during the charging process, and the discharging capacity D during the discharging process.
[0026] Next, step S20 is executed to construct a two-dimensional spatial image based on the time-series recording curve.
[0027] In some embodiments, the specific steps for constructing a two-dimensional spatial image based on the time-series recording curves are as follows: using a sliding window approach, the time-series data corresponding to each time-series recording curve is divided into multiple sub-sequences; the sub-sequences are mapped to vectors, and the distance between any two vectors is calculated to construct a distance matrix, where each element of the distance matrix reflects the difference in the electrochemical state of the supercapacitor within two specific time windows; the distance matrix is then transformed into the two-dimensional spatial image. By constructing a two-dimensional spatial image, a two-dimensional spatial representation of local time-series correlation features is achieved, which can fully leverage the advantages of convolutional neural networks in feature extraction, enabling the model to more sensitively capture nonlinear degradation patterns and improve the prediction accuracy of the remaining lifespan of the supercapacitor.
[0028] In some embodiments, before dividing the time series data corresponding to each time series recording curve into multiple subsequences using a sliding window method, the time series recording curve needs to be normalized in size. The vector corresponding to the time series recording curve is normalized to an integer multiple of one hundred data points by rounding down, and the data is normalized to the range of [0, 1].
[0029] Specifically, a sliding window with a fixed window size W can be used to move along the time-series recording curve with a step size of S, ultimately yielding N subsequences.
[0030] Where L is the length of the time-series recording curve after size normalization.
[0031] To visualize the time-series data, the subsequences can be mapped to vectors, and the Euclidean distance between any two vectors can be calculated using the following formula to construct the distance matrix.
[0032] in, For the first The vectors corresponding to each subsequence For the first The vectors corresponding to each subsequence For vectors , European distance, It can be viewed as an N*N two-dimensional spatial image.
[0033] When multiple time-series recording curves are included, the method in step S20 can be used to convert each time-series recording curve into a two-dimensional spatial image. For example, when the three time-series recording curves obtained in step S10 represent the complete voltage V during the charging and discharging process, the charging capacity C during the charging process, and the discharging capacity D during the discharging process, three two-dimensional spatial images can be constructed. Since the lengths of the three time-series recording curves—complete voltage V, charging capacity C, and discharging capacity D—are not exactly the same, the sizes of their corresponding two-dimensional spatial images are also not exactly the same.
[0034] Next, step S30 is performed to stack the channels based on the two-dimensional spatial image to generate a multi-channel fused image.
[0035] In some embodiments, in order to meet the input requirements of the convolutional neural network, the specific steps of channel stacking based on the two-dimensional spatial image to generate a multi-channel fused image are as follows: the two-dimensional spatial image is downsampled to scale its size to the size required by the convolutional neural network; the downsampled two-dimensional spatial image is stacked to generate a multi-channel fused image, thereby realizing a comprehensive characterization of the multi-dimensional electrochemical state characteristics of the supercapacitor.
[0036] It should be noted that, taking the generation of a three-channel fused image as an example: When a time-series recording curve is selected, the two-dimensional spatial image constructed by the time-series recording curve can be used as a single component of the multi-channel image. That is, copies of the grayscale image are filled into the three channels respectively to form a three-channel fused image. When two time-series recording curves are selected, the two-dimensional spatial images constructed by the two time-series recording curves can be filled into two channels of the multi-channel image respectively. The third channel uses a copy of the two-dimensional spatial image of one of the time-series recording curves or is filled as a blank channel, thereby forming a three-channel fused image. When three time-series recording curves are selected, the two-dimensional spatial images constructed by the three time-series recording curves can be filled into the three channels of the multi-channel image, thereby forming a three-channel fused image.
[0037] As an example, after constructing three two-dimensional spatial images in step S20—the complete voltage V during the charging and discharging process, the charging capacity C during the charging process, and the discharging capacity D during the discharging process—the three two-dimensional spatial images corresponding to the complete voltage curve during the charging and discharging process, the charging capacity curve during the charging process, and the discharging capacity curve during the discharging process can be stacked in channels to generate a three-channel fused image.
[0038] It should be noted that the image fusion in this application does not refer to pixel-level stitching in traditional computer vision, but rather to mapping one-dimensional time-series data into a two-dimensional spatial topology through specific mathematical transformations (such as distance matrices), and stacking the two-dimensional spatial structures corresponding to different physical quantities (such as voltage, charging capacity, and discharging capacity) as independent image channels in three-dimensional tensors. Its function is to present the nonlinear coupling relationship of multiple physical fields to the deep learning model in the form of multi-channel spatial features.
[0039] Figure 2 The diagram shows a schematic of a three-channel fused image constructed from three time-series recording curves extracted from the early five charge-discharge cycles of a certain supercapacitor cell: the complete voltage V during the charge-discharge process, the charging capacity C during the charging process, and the discharging capacity D during the discharging process.
[0040] Finally, step S40 is executed, in which the multi-channel fused image is input into a pre-trained convolutional neural network for feature extraction, and the remaining lifespan prediction result of the target supercapacitor is output.
[0041] In some embodiments, the convolutional neural network can be, for example, AlexNet, but other convolutional neural networks such as ResNet, DenseNet, VGG, and EfficientNet can also be used. AlexNet's input is a 224×224×3 image (or padded to 227×227×3) with RGB three channels. AlexNet contains 5 convolutional layers (including 3 pooling layers) and 3 fully connected layers. Each convolutional layer includes a convolutional kernel, a bias term, a ReLU activation function, and a local response normalization module. The 1st, 2nd, and 5th convolutional layers are followed by a max-pooling layer, and the last three layers are fully connected layers. The final output layer is a Softmax layer, which converts the network output into probability values used to predict the remaining lifespan of the target supercapacitor.
[0042] This application inputs the multi-channel fused image into a pre-trained convolutional neural network, which can automatically learn the cross-channel coupling relationships between these multiple physical quantities while performing feature extraction. This deep fusion of multimodal features greatly enriches the information entropy and significantly improves the model's generalization ability and prediction accuracy.
[0043] In some embodiments, the pre-trained convolutional neural network is trained as follows: multiple training samples are acquired, each training sample including a multi-channel fused sample image and a corresponding true cycle lifetime label; the multi-channel fused sample image is used as input, and the true cycle lifetime label is used as the label, to train the initial convolutional neural network model. The error gradient between the predicted value and the true label is continuously calculated using the backpropagation algorithm, and the network weights are updated accordingly.
[0044] The training samples can be constructed using publicly available supercapacitor datasets, for example. Specifically, the multi-channel fused sample images can be formed based on publicly available supercapacitor datasets, following the same steps as S10-S30 described above; the cycle life label is selected based on the number of complete charge-discharge cycles that the supercapacitor can withstand before its capacity or equivalent series resistance, or other key performance parameters, decays to the failure threshold.
[0045] As an example, cycle life can be defined as the number of complete charge-discharge cycles a supercapacitor can withstand before its capacity decays to 80% (configurable as needed) of its initial capacity; or cycle life can be defined as the number of complete charge-discharge cycles a supercapacitor can withstand before its equivalent series resistance decays to twice its initial internal resistance (configurable as needed).
[0046] Based on the same concept, such as Figure 3As shown, this application also provides a supercapacitor remaining lifespan prediction device 11, including a data acquisition module 111, an image construction module 112, an image fusion module 113, and a lifespan prediction module 114.
[0047] Among them, the data acquisition module 111 is used to acquire at least one time-series recording curve of the target supercapacitor during the early charge-discharge cycle process; Image construction module 112 is used to construct a two-dimensional spatial image based on the time-series recording curve; Image fusion module 113 is used to stack channels according to the two-dimensional spatial image to generate a multi-channel fused image; The lifespan prediction module 114 is used to input the multi-channel fused image into a pre-trained convolutional neural network for feature extraction and output the remaining lifespan prediction result of the target supercapacitor.
[0048] It should be noted that the supercapacitor remaining lifespan prediction device 11 provided in the above embodiments and the supercapacitor remaining lifespan prediction method provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments, and will not be repeated here. In practical applications, the supercapacitor remaining lifespan prediction device 11 provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.
[0049] In summary, the supercapacitor remaining lifespan prediction method and apparatus of this application includes acquiring at least one time-series recording curve of a target supercapacitor during its early charge-discharge cycle; constructing a two-dimensional spatial image based on the time-series recording curve; stacking channels of the two-dimensional spatial image to generate a multi-channel fused image; inputting the multi-channel fused image into a pre-trained convolutional neural network for feature extraction; and outputting the predicted remaining lifespan of the target supercapacitor. By extracting the time-series recording curve of the supercapacitor during its early charge-discharge cycle to predict the remaining lifespan of the supercapacitor, the method solves the problem of prediction lag caused by the long-term data accumulation required by traditional methods. By constructing the time-series recording curve into a multi-channel fused image and inputting it into a convolutional neural network, the method fully leverages the advantages of convolutional neural networks in feature extraction, deeply mining the local temporal correlation and multi-channel coupling features in the image, improving the prediction accuracy of the supercapacitor's remaining lifespan, thereby achieving early, rapid, and high-precision prediction of the supercapacitor's remaining lifespan, providing support for the screening and optimized control of energy storage systems.
[0050] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.
[0051] The above description of the embodiments shown in this application (including the content set forth in the abstract of the specification) is not intended to be an exhaustive enumeration or to limit this application to the precise forms disclosed herein. Although specific embodiments and examples of this application have been described herein for illustrative purposes only, various equivalent modifications are possible within the spirit and scope of this application, as will be recognized and understood by those skilled in the art. As indicated, these modifications can be made to this application in accordance with the above description of the embodiments described herein, and such modifications will be within the spirit and scope of this application.
[0052] Therefore, although this application has been described herein with reference to specific embodiments thereof, freedom of modification, various changes and substitutions are also within the scope of the above disclosure, and it should be understood that in some cases, certain features of this application may be adopted without departing from the scope and spirit of the application and without corresponding use of other features. Thus, many modifications can be made to adapt a particular environment or material to the substantive scope and spirit of this application. This application is not intended to be limited to the specific terminology used in the following claims and / or the specific embodiments disclosed as the best mode contemplated for carrying out this application, but this application will include any and all embodiments and equivalents falling within the scope of the appended claims. Therefore, the scope of this application will be determined only by the appended claims.
Claims
1. A method of predicting the remaining useful life of a supercapacitor, characterized in that, include: Obtain at least one time-series recording curve of the target supercapacitor during the early charge-discharge cycle process; A two-dimensional spatial image is constructed based on the time-series recording curve; Channel stacking is performed on the two-dimensional spatial image to generate a multi-channel fused image; The multi-channel fused image is input into a pre-trained convolutional neural network for feature extraction, and the remaining lifespan prediction result of the target supercapacitor is output.
2. The supercapacitor remaining useful lifetime prediction method of claim 1, wherein, Based on the time-series recording curve, a two-dimensional spatial image is constructed, including: A sliding window approach is used to divide the time series data corresponding to each time series record curve into multiple subsequences; The subsequences are mapped to vectors, and the distance between any two vectors is calculated to construct a distance matrix; The distance matrix is converted into the two-dimensional spatial image.
3. The method of claim 2, wherein, Before using a sliding window approach to divide the time series data corresponding to each time series record curve into multiple subsequences, the method further includes: The time-series recording curve is normalized in size, and the vector corresponding to the time-series recording curve is normalized to an integer multiple of one hundred data points, and the data is normalized to the range of [0,1].
4. The method of claim 2, wherein, Mapping the subsequences to vectors and calculating the distance between any two vectors to construct a distance matrix includes: The subsequences are mapped to vectors, and the Euclidean distance between any two vectors is calculated to construct the distance matrix.
5. The method for predicting the remaining service life of a supercapacitor according to claim 2, characterized in that, The convolutional neural network includes AlexNet.
6. The method for predicting the remaining service life of a supercapacitor according to claim 1, characterized in that, Channel stacking is performed on the two-dimensional spatial image to generate a multi-channel fused image, including: The two-dimensional spatial image is downsampled to scale its size to the size required by the convolutional neural network. The downsampled two-dimensional spatial image is stacked into channels to generate a multi-channel fused image.
7. The method for predicting the remaining service life of a supercapacitor according to claim 1, characterized in that, The acquisition of at least one time-series recording curve of the target supercapacitor during the early charge-discharge cycle includes: Obtain three time-series recording curves of the target supercapacitor during several early charge-discharge cycles, wherein the three time-series recording curves include a complete voltage curve during the charge-discharge process, a charging capacity curve during the charging process, and a discharging capacity curve during the discharging process.
8. The method for predicting the remaining service life of a supercapacitor according to claim 7, characterized in that, Channel stacking is performed on the two-dimensional spatial image to generate a multi-channel fused image, including: The three two-dimensional spatial images corresponding to the complete voltage curve during the charging and discharging process, the charging capacity curve during the charging process, and the discharging capacity curve during the discharging process are stacked in channels to generate a three-channel fused image.
9. The method for predicting the remaining service life of a supercapacitor according to claim 1, characterized in that, The pre-trained convolutional neural network is trained in the following way: Multiple training samples are acquired, each training sample including a multi-channel fused sample image and a corresponding true cycle lifetime label; The multi-channel fused sample image is used as input, and the real cycle lifetime label is used as the label to train the initial convolutional neural network model.
10. A device for predicting the remaining service life of a supercapacitor, characterized in that, include: The data acquisition module is used to acquire at least one time-series recording curve of the target supercapacitor during the early charge-discharge cycle process; The image construction module is used to construct a two-dimensional spatial image based on the time-series recording curve; An image fusion module is used to stack channels based on the two-dimensional spatial image to generate a multi-channel fused image; The lifespan prediction module is used to input the multi-channel fused image into a pre-trained convolutional neural network for feature extraction and output the remaining lifespan prediction result of the target supercapacitor.