Capacitor life prediction method and system based on capacitance attenuation rate

Abstract

The application discloses a capacitor life prediction method and system based on a capacitor attenuation rate, and the method comprises capacitor operation data collection, out-of-group passivation loss construction, time-series capacitor attenuation loss construction, attenuation constraint loss construction, time-varying coefficient calculation, capacitor life prediction model design and capacitor life prediction. The application belongs to the field of data processing, and specifically refers to a capacitor life prediction method and system based on a capacitor attenuation rate. The scheme is designed for the time-varying out-of-group passivation loss of the out-of-group and noise problems, suppresses the out-of-group points, avoids the overall deviation of the life prediction, constructs the time-varying scale parameter to remove the false micro shaking, constructs the time-series capacitor attenuation loss fused with acceleration sensing, introduces the second-order difference of the capacitor attenuation rate to capture the accelerated degradation characteristics, constructs the attenuation constraint loss to improve the rationality of the long-term life prediction, constructs the time-varying coefficient to realize dynamic weight distribution, realizes the near short-term high-precision and long-term stable trend, and improves the capacitor life prediction effect.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for predicting capacitor lifespan based on capacitance decay rate. Background Technology

[0002] Capacitor life prediction methods monitor operating data such as capacitance decay rate, capacitance, temperature, and voltage, and use data fitting or machine learning models to fit the aging trend, extrapolating the remaining time before performance degrades to the failure threshold. However, general capacitor life prediction methods are sensitive to noise and outliers, have a weak ability to capture abrupt changes in the later stages of aging, and thus suffer from life prediction bias. Furthermore, general capacitor life prediction methods exhibit large long-term life prediction deviations and are prone to unreasonable drift, while short-term predictions are not precise enough. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a capacitor lifetime prediction method and system based on capacitance decay rate. Addressing the problems of general capacitor lifetime prediction methods being sensitive to noise and outliers, and having weak ability to capture abrupt changes in late-stage aging, leading to lifetime prediction bias, this solution designs a time-varying outlier passivation loss to suppress random outliers caused by measurement noise and electromagnetic interference, thus avoiding overall lifetime prediction bias. It constructs time-varying scale parameters, using high sensitivity in the early stages to eliminate spurious micro-jitter, and retaining true large-scale decay abrupt changes in the later stages. It also constructs a time-series capacitance decay loss fused with acceleration sensing: the capacitance decay rate residual is time-weighted, and an exponential time decay term is used to improve the... This approach weights recent data to reduce the impact of long-term data; it introduces the second-order difference of capacitor decay rate as a characterization of aging acceleration, accurately capturing accelerated degradation characteristics; thus improving the accuracy of capacitor life prediction. Addressing the issues of large long-term life prediction deviations and unreasonable drift in general capacitor life prediction methods, while short-term predictions are not precise enough, this scheme constructs a decay constraint loss, using the difference between theoretical energy storage and energy storage calculated from measured capacitance and voltage to build a physical constraint term, improving the rationality and stability of long-term life prediction; it constructs a time-varying coefficient that depends on both the remaining life prediction duration and the real-time capacitor decay rate prediction error, achieving dynamic weight allocation, high accuracy in the near and short term, and stable trends in the long term; thus improving the performance of capacitor life prediction.

[0004] The technical solution adopted by this invention is as follows: This invention provides a capacitor lifetime prediction method based on capacitance decay rate, the method comprising the following steps: Step S1: Capacitor operation data acquisition; Step S2: Constructing the outlier passivation loss; Step S3: Constructing the timing capacitance decay loss; Step S4: Constructing the decay constraint loss; Step S5: Calculation of time-varying coefficients; Step S6: Design of capacitor lifetime prediction model; Step S7: Capacitor life prediction.

[0005] Furthermore, in step S1, the capacitor operation data acquisition involves obtaining historical operating sequence data of the capacitor under actual operating conditions, labeling the capacitor attenuation rate, performing preprocessing, and finally constructing a standardized dataset.

[0006] Furthermore, in step S2, the outlier passivation loss construction involves constructing an outlier passivation loss function to perform time-varying passivation on the capacitor attenuation rate error, adjusting the width parameter according to the real-time health status to obtain the outlier passivation loss function.

[0007] Furthermore, in step S3, the construction of the time-series capacitance decay loss is to construct a time-series capacitance decay loss that integrates decay acceleration perception. This not only assigns higher weights to sampling points closer to the current time, but also introduces second-order difference features as weight enhancement factors to obtain the time-series capacitance decay loss.

[0008] Furthermore, in step S4, the attenuation constraint loss is constructed by embedding the energy conservation characteristics of the electrochemical attenuation process of the capacitor into the loss function to construct the attenuation constraint loss.

[0009] Furthermore, in step S5, the time-varying coefficient calculation is to design a time-varying coefficient that simultaneously depends on the capacitor decay rate prediction error and the future remaining lifetime prediction duration.

[0010] Further, in step S6, the capacitor lifetime prediction model design specifically includes: The model architecture is designed as follows: an input layer, which takes a standardized dataset as input; a temporal encoding layer, which performs nonlinear mapping and temporal correlation processing on the input features to extract shallow aging features; a shared feature extraction layer, which performs deep transformation and correlation mining on the encoded features to output deep shared features; a dual-branch prediction output layer, where branch 1 predicts the evolution trajectory of capacitance decay rate at future times, and branch 2 directly infers the remaining service life based on the evolution trajectory; and a time-varying constraint optimization layer, which calculates the time-varying coefficients, temporal capacitance decay loss, and decay constraint loss, fuses them to obtain the total loss, and backpropagates to optimize the model parameters. The shared feature extraction layer definition performs nonlinear transformation and temporal correlation modeling on the input features to capture the degradation law of capacitor decay rate with operating conditions and time, providing deep features for subsequent bi-branch prediction; The dual-branch prediction layer design consists of two branches: Branch 1: Capacitor decay rate prediction branch, which predicts the capacitor decay rate at consecutive future moments; and Branch 2: Remaining lifetime prediction branch, which obtains the remaining lifetime based on the future decay rate evolution trend and by mapping the failure threshold. The total loss function is designed by fusing the time-series capacitance decay loss and decay constraint loss through time-varying coefficients to construct a unified training objective.

[0011] Furthermore, in step S7, the capacitor lifetime prediction involves real-time acquisition of the capacitor's runtime sequence data, which is then preprocessed and input into the trained capacitor lifetime prediction model. The model sequentially performs capacitor attenuation rate prediction and remaining lifetime inference, outputting the real-time remaining lifespan of the capacitor at future moments, thereby realizing capacitor lifetime prediction.

[0012] The present invention provides a capacitor lifetime prediction system based on capacitance decay rate, including a capacitor operation data acquisition module, an outlier passivation loss construction module, a time-series capacitance decay loss construction module, a decay constraint loss construction module, a time-varying coefficient calculation module, a capacitor lifetime prediction model design module, and a capacitor lifetime prediction module. The capacitor operation data acquisition module acquires the historical operation sequence data of the capacitor, extracts the capacitance decay rate, performs preprocessing, and finally constructs a standardized dataset. The outlier passivation loss construction module constructs an outlier passivation loss function with time-varying scale parameters to perform time-varying passivation processing on the capacitance attenuation rate error; The time-series capacitance decay loss construction module combines the time decay characteristics and the second-order difference characteristics of capacitance decay rate to construct a time-series capacitance decay loss that integrates decay acceleration perception. The attenuation constraint loss construction module constructs an attenuation constraint loss function based on capacitor electrochemical attenuation. The time-varying coefficient calculation module is designed to rely on the time-varying coefficients of the capacitor decay rate prediction error and the remaining lifetime prediction duration. The capacitor lifetime prediction model design module constructs a capacitor lifetime prediction model on a standardized dataset based on outlier passivation loss, time-series capacitance decay loss, decay constraint loss, and time-varying coefficients. The capacitor life prediction module predicts the life of capacitors based on capacitor life prediction models and operating data.

[0013] The beneficial effects achieved by the present invention using the above solution are as follows: (1) To address the problem that general capacitor lifetime prediction methods are sensitive to noise and outliers and have a weak ability to capture abrupt changes in the later stages of aging, which leads to a shift in lifetime prediction, this scheme designs a time-varying outlier passivation loss to suppress random outliers caused by measurement noise and electromagnetic interference, thus avoiding an overall shift in lifetime prediction; it constructs time-varying scale parameters to remove false small jitters with high sensitivity in the early stage and retain real large attenuation abrupt changes in the later stage; it constructs a time-series capacitor attenuation loss that integrates acceleration perception: it performs time-series weighting on the capacitor attenuation rate residual, increases the weight of recent data through the exponential time attenuation term, and reduces the influence of long-term data; it introduces the second-order difference of the capacitor attenuation rate as a characterization of aging acceleration to accurately capture accelerated degradation characteristics; thus improving the accuracy of capacitor lifetime prediction.

[0014] (2) In view of the problems that general capacitor life prediction methods have large long-term life prediction deviations and are prone to unreasonable drift, and short-term predictions are not precise enough, this scheme constructs a decay constraint loss, constructs a physical constraint term based on the difference between theoretical energy storage and energy storage calculated by measured capacitance and voltage, and improves the rationality and stability of long-term life prediction; constructs a time-varying coefficient that depends on both the remaining life prediction duration and the real-time capacitance decay rate prediction error, realizes dynamic weight allocation, high accuracy in the near and short term and stable trend in the long term; thereby improving the capacitor life prediction effect. Attached Figure Description

[0015] Figure 1 A schematic flowchart of a capacitor lifetime prediction method based on capacitance decay rate provided by the present invention; Figure 2 This is a schematic diagram of a capacitor lifetime prediction system based on capacitance decay rate provided by the present invention.

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0019] Example 1, see Figure 1 This invention provides a capacitor lifetime prediction method based on capacitance decay rate, the method comprising the following steps: Step S1: Capacitor operation data acquisition, obtain historical operation sequence data of capacitor, extract capacitance decay rate, perform preprocessing, and finally construct a standardized dataset; Step S2: Outlier passivation loss construction. Construct an outlier passivation loss function with time-varying scale parameters to perform time-varying passivation processing on the capacitance attenuation rate error. Step S3: Constructing the time-series capacitance decay loss. Combining the time decay characteristics and the second-order difference characteristics of the capacitance decay rate, construct a time-series capacitance decay loss that integrates decay acceleration perception. Step S4: Construction of attenuation constraint loss, constructing an attenuation constraint loss function based on capacitor electrochemical attenuation; Step S5: Calculate the time-varying coefficients and design time-varying coefficients that depend on the prediction error of the capacitor decay rate and the prediction duration of the remaining lifetime; Step S6: Design of capacitor lifetime prediction model. Based on outlier passivation loss, time-series capacitance decay loss, decay constraint loss and time-varying coefficient, a capacitor lifetime prediction model is constructed on a standardized dataset. Step S7: Capacitor life prediction. Based on the capacitor life prediction model, the life of the capacitor is predicted from the operating data.

[0020] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the capacitor operation data acquisition involves obtaining historical operation sequence data of the capacitor under actual working conditions, extracting the core health characteristic of capacitance decay rate, and performing preprocessing to finally construct a standardized dataset. The acquisition includes time-series operation data of the capacitor under actual working conditions, including real-time capacitance value, terminal voltage, ambient temperature, charging and discharging current, number of charging and discharging cycles, and cumulative operating time. The capacitance decay rate is calculated based on the measured capacity and the initial rated capacity to characterize the aging degree of the capacitor. Subsequently, the original data undergoes preprocessing operations such as missing value imputation, preliminary outlier removal, and maximum-minimum normalization.

[0021] Example 3, see Figure 1This embodiment is based on the above embodiment. In step S2, the outlier passivation loss construction addresses the problem that the original monitoring data is affected by the accuracy of the measuring equipment and on-site electromagnetic interference, resulting in random noise and extreme outlier mutation points. To avoid lifetime prediction deviation, an outlier passivation loss function is constructed to perform time-varying passivation on the capacitance attenuation rate error. The width parameter is adjusted time-varyingly according to the real-time health status, expressed as: ; ;in, It is the outlier passivation loss function; It is the predicted residual of the capacitance decay rate at time k. ; This is the predicted value of the capacitance decay rate at the corresponding time, where k is the time index and N is the total number of times. This is the measured value of the capacitance decay rate at the corresponding moment; It is a time-varying scaling parameter; This is the baseline noise level; It is a degradation sensitivity modulator, with a value ranging from 2.0 to 5.0; It represents the healthy state of the capacitor at time k. , It is the measured capacitance value at time k. This is the rated capacitance value of the capacitor; Maintaining high sensitivity in the early stages to eliminate minor jitters, and automatically relaxing the penalty for large amplitude values ​​in the later stages, thereby preserving true pre-failure characteristics and avoiding misjudging fault signals as noise and eliminating them, significantly improving the ability to capture abrupt changes in the later stages of aging from a data perspective.

[0022] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the construction of the time-series capacitance decay loss is based on the fact that in the aging law of capacitors, the future remaining lifetime is highly dependent on the evolution direction and acceleration of the capacitance decay trend. The contribution of the recent capacitance decay trend to the judgment of the remaining lifetime is significantly higher than that of the long-term historical data. Therefore, a time-series capacitance decay loss that integrates decay acceleration perception is constructed. Not only are sampling points closer to the current time given higher weights, but second-order difference features are also introduced as weight enhancement factors. The aim is to automatically identify and lock the critical region where the capacitor transitions from linear aging to accelerated aging, expressed as: ; ; ;in, This is the loss due to the decay of the timing capacitor; It is the coefficient at time k; It is the second-order difference of the capacitance decay rate, which characterizes the acceleration of the aging process. The second-order difference in the first two moments uses a one-sided difference. It is the acceleration gain factor, with a value ranging from 0.5 to 2.0; It is the cumulative running time of the capacitor at time k; It is the time decay adjustment factor, with a value of 0.005 to 0.05.

[0023] By performing the above operations, this scheme addresses the problems of general capacitor lifetime prediction methods, such as sensitivity to noise and outliers, weak ability to capture abrupt changes in late-stage aging, and consequently, lifetime prediction bias. It designs a time-varying outlier passivation loss to suppress random outliers caused by measurement noise and electromagnetic interference, thus avoiding overall lifetime prediction bias. A time-varying scale parameter is constructed to remove false small fluctuations with high sensitivity in the early stages and retain true large-scale decay abrupt changes in the later stages. A time-series capacitor decay loss based on acceleration perception is constructed: the capacitor decay rate residual is time-weighted, and the weight of recent data is increased through an exponential time decay term to reduce the influence of long-term data. The second-order difference of the capacitor decay rate is introduced as a characterization of aging acceleration, accurately capturing accelerated degradation characteristics. These measures ultimately improve the accuracy of capacitor lifetime prediction.

[0024] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the attenuation constraint loss construction is problematic because pure data models are prone to unreasonable drift phenomena such as non-monotonic predicted attenuation rates and energy storage changes violating physical laws, especially in long-term lifetime predictions where deviations are significant. Therefore, the energy conservation characteristics of the capacitor's electrochemical attenuation process are embedded into the loss function to construct the attenuation constraint loss, forcing the model output to conform to the actual aging mechanism and avoiding unreasonable drift in long-term lifetime predictions. This is expressed as follows: ;in, It is a decay constraint loss; It is the theoretical energy stored in the capacitor at time k. It is the measured capacitance value of the capacitor at time k. It is the measured value of the capacitor terminal voltage at time k.

[0025] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the calculation of the time-varying coefficient is to achieve a time-varying balance between the historical data fitting accuracy and the strength of the future evolution mechanism regularization. It designs a time-varying coefficient that simultaneously depends on the capacitance decay rate prediction error and the future remaining lifetime prediction duration. When the prediction error is large or the prediction duration is far in the future, the data fitting weight is automatically reduced and the mechanism regularization weight is increased to avoid the model overfitting noise and violating the aging mechanism. When the prediction error is small and the prediction duration is recent, the data fitting weight is increased and the mechanism regularization is weakened to ensure short-term prediction accuracy, as expressed as: ;in, It is a time-varying coefficient, where t is the predicted remaining lifespan of the capacitor. This is the error in real-time prediction of capacitor attenuation rate. ; It is the error attenuation adjustment factor, with a value ranging from 0.1 to 1.0; It is the time decay adjustment factor, with a value ranging from 0.01 to 0.1; It is a time-varying lower limit, with a value of 0.1 to 0.2.

[0026] By performing the above operations, this scheme addresses the problems of large long-term life prediction deviations and unreasonable drift in general capacitor life prediction methods, as well as insufficient precision in short-term predictions. It constructs a decay constraint loss, using the difference between theoretical energy storage and energy storage calculated from measured capacitance and voltage to create a physical constraint term, thereby improving the rationality and stability of long-term life prediction. Furthermore, it constructs a time-varying coefficient that depends on both the remaining life prediction duration and the real-time capacitance decay rate prediction error, achieving dynamic weight allocation, high accuracy in the near and short term, and stable trends in the long and long term; thus improving the overall performance of capacitor life prediction.

[0027] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the capacitor lifetime prediction model design adopts an end-to-end architecture of operating condition feature input + time-series feature encoding + shared feature extraction + dual-branch output + time-varying mechanism regularization, specifically including: The model architecture is designed as follows: an input layer, which takes a standardized dataset as input; a temporal encoding layer, which performs nonlinear mapping and temporal correlation processing on the input features to extract shallow aging features; a shared feature extraction layer, which performs deep transformation and correlation mining on the encoded features to capture the complex degradation patterns between operating conditions, time, and capacitor decay, and outputs deep shared features with high representational capabilities; a dual-branch prediction output layer, where branch 1 predicts the evolution trajectory of capacitor decay rate at future times, and branch 2 directly infers the remaining service life based on the evolution trajectory; and a time-varying constraint optimization layer, which calculates the time-varying coefficients, temporal capacitor decay loss, and decay constraint loss, fuses them to obtain the total loss, and backpropagates to optimize the model parameters. The shared feature extraction layer is defined to perform nonlinear transformation and temporal correlation modeling on the input features, capturing the complex degradation law of capacitor attenuation rate as it changes with operating conditions and time. This provides high-representational-capability deep features for subsequent bi-branch prediction, expressed as: Where H is the deep shared aging feature vector extracted from the standardized dataset; It is a non-linear feature coding network, and LSTM is selected; It is the input feature vector at time k; The dual-branch prediction layer design includes branch 1, which predicts the capacitance decay rate at consecutive future time points, and is represented as follows: Branch 2: Remaining lifetime prediction branch. Based on the future decay rate evolution trend, the remaining lifetime is obtained by combining the failure threshold mapping, and is expressed as: ;in, This is the predicted capacitance decay rate; It is a capacitor decay rate prediction subnetwork, which uses a nonlinear regressor composed of two fully connected neural networks. The input is a shared feature vector H. The feature transformation is completed through the activation function, and the output is the capacitor decay rate at future time, so as to accurately characterize the aging trend. This is the predicted remaining lifespan of the capacitor; This is the capacitor failure threshold, set to 0.8; It is the decay rate, the amount of change in the rate of capacitance decay per unit time. The overall loss function design integrates the time-series capacitance decay loss and decay constraint loss through time-varying coefficients to construct a unified training objective. The larger the prediction error and the further back in the prediction time, the higher the proportion of mechanistic regularization, thereby ensuring accuracy while avoiding non-physical drift. This is expressed as: ;in, It is the total optimization loss of the model; Regarding model training: During model training, with the goal of minimizing the total optimization loss, an adaptive moment estimation optimizer is used to iteratively update the network parameters. In each iteration, the time-series capacitance decay loss, decay constraint loss, and time-varying coefficients are calculated simultaneously. The parameters of the shared feature extraction layer and the dual-branch prediction sub-network are continuously adjusted through backpropagation until the loss converges, ultimately obtaining a capacitor lifetime prediction model with stable prediction capabilities. The data directly involved in the calculation in the scheme are all normalized.

[0028] Example 8, see Figure 1 This embodiment is based on the above embodiment. In step S7, the capacitor lifetime prediction is to collect the capacitor's runtime sequence data in real time, and after preprocessing, input it into the trained capacitor lifetime prediction model. The model sequentially completes the capacitor decay rate prediction and remaining lifetime inference, and directly outputs the real-time remaining lifespan of the capacitor at future moments, thereby realizing capacitor lifetime prediction.

[0029] Example 9, see Figure 2 Based on the above embodiments, this embodiment provides a capacitor lifetime prediction system based on capacitance decay rate, including a capacitor operation data acquisition module, an outlier passivation loss construction module, a time-series capacitance decay loss construction module, a decay constraint loss construction module, a time-varying coefficient calculation module, a capacitor lifetime prediction model design module, and a capacitor lifetime prediction module. The capacitor operation data acquisition module acquires the historical operation sequence data of the capacitor, extracts the capacitance decay rate, performs preprocessing, and finally constructs a standardized dataset. The outlier passivation loss construction module constructs an outlier passivation loss function with time-varying scale parameters to perform time-varying passivation processing on the capacitance attenuation rate error; The time-series capacitance decay loss construction module combines the time decay characteristics and the second-order difference characteristics of capacitance decay rate to construct a time-series capacitance decay loss that integrates decay acceleration perception. The attenuation constraint loss construction module constructs an attenuation constraint loss function based on capacitor electrochemical attenuation. The time-varying coefficient calculation module is designed to rely on the time-varying coefficients of the capacitor decay rate prediction error and the remaining lifetime prediction duration. The capacitor lifetime prediction model design module constructs a capacitor lifetime prediction model on a standardized dataset based on outlier passivation loss, time-series capacitance decay loss, decay constraint loss, and time-varying coefficients. The capacitor life prediction module predicts the life of capacitors based on capacitor life prediction models and operating data.

[0030] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0031] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0032] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for predicting capacitor lifetime based on capacitance decay rate, characterized in that: The method includes the following steps: Step S1: Capacitor operation data acquisition, obtain historical operation sequence data of capacitor, extract capacitance decay rate, perform preprocessing, and finally construct a standardized dataset; Step S2: Outlier passivation loss construction. Construct an outlier passivation loss function with time-varying scale parameters to perform time-varying passivation processing on the capacitance attenuation rate error. Step S3: Constructing the time-series capacitance decay loss. Combining the time decay characteristics and the second-order difference characteristics of the capacitance decay rate, construct a time-series capacitance decay loss that integrates decay acceleration perception. Step S4: Construction of attenuation constraint loss, constructing an attenuation constraint loss function based on capacitor electrochemical attenuation; Step S5: Calculate the time-varying coefficients and design time-varying coefficients that depend on the prediction error of the capacitor decay rate and the prediction duration of the remaining lifetime; Step S6: Design of capacitor lifetime prediction model. Based on outlier passivation loss, time-series capacitance decay loss, decay constraint loss and time-varying coefficient, a capacitor lifetime prediction model is constructed on a standardized dataset. Step S7: Capacitor life prediction. Based on the capacitor life prediction model, the life of the capacitor is predicted from the operating data.

2. The capacitor lifetime prediction method based on capacitance decay rate according to claim 1, characterized in that: In step S2, the outlier passivation loss construction involves constructing an outlier passivation loss function to perform time-varying passivation on the capacitor attenuation rate error. The width parameter is adjusted time-varyingly according to the real-time health status to obtain the outlier passivation loss function.

3. The capacitor lifetime prediction method based on capacitance decay rate according to claim 1, characterized in that: In step S3, the construction of the time-series capacitance decay loss is to construct a time-series capacitance decay loss that integrates decay acceleration sensing. This not only assigns higher weights to sampling points closer to the current time, but also introduces second-order difference features as weight enhancement factors to obtain the time-series capacitance decay loss.

4. The capacitor lifetime prediction method based on capacitance decay rate according to claim 1, characterized in that: In step S4, the decay constraint loss is constructed by embedding the energy conservation characteristics of the electrochemical decay process of the capacitor into the loss function to construct the decay constraint loss.

5. The capacitor lifetime prediction method based on capacitance decay rate according to claim 1, characterized in that: In step S5, the time-varying coefficient calculation is to design a time-varying coefficient that simultaneously depends on the capacitor decay rate prediction error and the future remaining lifetime prediction duration.

6. The capacitor lifetime prediction method based on capacitance decay rate according to claim 1, characterized in that: In step S6, the capacitor lifetime prediction model design specifically includes: The model architecture is designed as follows: an input layer, which takes a standardized dataset as input; a temporal encoding layer, which performs nonlinear mapping and temporal correlation processing on the input features to extract shallow aging features; a shared feature extraction layer, which performs deep transformation and correlation mining on the encoded features to output deep shared features; a dual-branch prediction output layer, where branch 1 predicts the evolution trajectory of capacitance decay rate at future times, and branch 2 directly infers the remaining service life based on the evolution trajectory; and a time-varying constraint optimization layer, which calculates the time-varying coefficients, temporal capacitance decay loss, and decay constraint loss, fuses them to obtain the total loss, and backpropagates to optimize the model parameters. The shared feature extraction layer definition performs nonlinear transformation and temporal correlation modeling on the input features to capture the degradation law of capacitor decay rate with operating conditions and time, providing deep features for subsequent bi-branch prediction; The dual-branch prediction layer design consists of two branches: Branch 1: Capacitor decay rate prediction branch, which predicts the capacitor decay rate at consecutive future moments; and Branch 2: Remaining lifetime prediction branch, which obtains the remaining lifetime based on the future decay rate evolution trend and by mapping the failure threshold. The total loss function is designed by fusing the time-series capacitance decay loss and decay constraint loss through time-varying coefficients to construct a unified training objective.

7. The capacitor lifetime prediction method based on capacitance decay rate according to claim 6, characterized in that: In step S7, the capacitor lifetime prediction involves real-time acquisition of the capacitor's runtime sequence data, which is then preprocessed and input into the trained capacitor lifetime prediction model. The model sequentially performs capacitor attenuation rate prediction and remaining lifetime inference, outputting the capacitor's real-time remaining lifetime at future moments, thereby achieving capacitor lifetime prediction.

8. A capacitor lifetime prediction system based on capacitance decay rate, used to implement the capacitor lifetime prediction method based on capacitance decay rate as described in any one of claims 1-7, characterized in that: It includes a capacitor operation data acquisition module, an outlier passivation loss construction module, a time-series capacitance decay loss construction module, a decay constraint loss construction module, a time-varying coefficient calculation module, a capacitor lifetime prediction model design module, and a capacitor lifetime prediction module; The capacitor operation data acquisition module acquires the historical operation sequence data of the capacitor, extracts the capacitance decay rate, performs preprocessing, and finally constructs a standardized dataset. The outlier passivation loss construction module constructs an outlier passivation loss function with time-varying scale parameters to perform time-varying passivation processing on the capacitance attenuation rate error; The time-series capacitance decay loss construction module combines the time decay characteristics and the second-order difference characteristics of capacitance decay rate to construct a time-series capacitance decay loss that integrates decay acceleration perception. The attenuation constraint loss construction module constructs an attenuation constraint loss function based on capacitor electrochemical attenuation. The time-varying coefficient calculation module is designed to rely on the time-varying coefficients of the capacitor decay rate prediction error and the remaining lifetime prediction duration. The capacitor lifetime prediction model design module constructs a capacitor lifetime prediction model on a standardized dataset based on outlier passivation loss, time-series capacitance decay loss, decay constraint loss, and time-varying coefficients. The capacitor life prediction module predicts the life of capacitors based on capacitor life prediction models and operating data.

Images