CN-121996980-A - Proton exchange membrane fuel cell degradation trend prediction method based on frequency enhancement direct prediction

CN121996980ACN 121996980 ACN121996980 ACN 121996980ACN-121996980-A

Abstract

The invention discloses a proton exchange membrane fuel cell degradation trend prediction method based on frequency enhancement direct prediction, which belongs to the field of fuel cell life prediction and comprises the steps of obtaining fuel cell voltage time sequence aging data and preprocessing to construct a sample set, constructing a ITransformer architecture-based sequence-to-sequence prediction model, adopting a frequency enhancement direct prediction training paradigm training model, fusing loss functions of a time domain and a frequency domain loss component, adopting Bayesian optimization and cross-validation to conduct super-parameter optimization to determine a final model, and predicting the voltage degradation trend by using the model. According to the invention, the multi-step prediction precision is improved through the frequency domain loss constraint, the parameter optimizing efficiency and stability are improved through the intelligent optimizing strategy, and the robustness of the model to actual noise is enhanced through targeted preprocessing, so that the more accurate and efficient prediction of the degradation trend of the proton exchange membrane fuel cell is realized.

Inventors

Li Mince
Pei Tianyin
PAN TIANHONG
TIAN JIAQIANG
FANG LE
ZHANG DEXIANG
FAN YUAN

Assignees

安徽大学

Dates

Publication Date: 20260508
Application Date: 20260205

Claims (10)

1. The method for predicting the degradation trend of the proton exchange membrane fuel cell based on the frequency enhancement direct prediction is characterized by comprising the following steps of: s1, acquiring voltage time sequence aging data of a proton exchange membrane fuel cell, and preprocessing the voltage time sequence aging data to construct an input sample set; s2, constructing a ITransformer architecture-based sequence-to-sequence prediction model; S3, training the sequence-to-sequence prediction model by adopting a frequency enhancement direct prediction training paradigm, wherein the total loss function of the frequency enhancement direct prediction training paradigm is formed by weighted summation of a time domain loss component and a frequency domain loss component; S4, performing super-parameter optimization on the trained sequence-to-sequence prediction model by adopting a Bayesian optimization method, and evaluating the model performance by adopting a cross verification method to determine a final prediction model; s5, predicting the voltage degradation trend of the proton exchange membrane fuel cell by using the final prediction model.
2. The prediction method according to claim 1, wherein in step S1, the preprocessing includes denoising the voltage time-series aging data using a variational modal decomposition method, and processing the denoised time series using a sliding window method to construct the input sample set, wherein an input time length of each sample is greater than an output prediction time length.
3. The prediction method according to claim 1, wherein in step S2, the input of the sequence to the sequence prediction model is a voltage history sequence subjected to an instance normalization process.
4. The prediction method according to claim 1, wherein in step S3, the frequency domain loss component is calculated by converting the predicted sequence and the true sequence to the frequency domain by fourier transform, respectively, and calculating the L1 norm difference between the two in the frequency domain.
5. The prediction method according to claim 4, wherein in step S3, the formula of the frequency domain loss component is: ; Wherein, the For the frequency domain lost component, In order to be a true sequence of the sequences, Is a predicted sequence.
6. The prediction method according to claim 4, wherein in step S3, the total loss function: ; Wherein, the For the time-domain loss component, Is a preset weight coefficient.
7. The prediction method according to claim 1, wherein in step S4, the bayesian optimization method is used for performing an optimization search on at least one super parameter of an embedding dimension, an encoder layer number and an attention header number of the sequence-to-sequence prediction model.
8. The method according to claim 7, wherein in step S4, the cross-validation method is K-fold cross-validation, wherein K is 5 or more.
9. The prediction method according to claim 8, wherein in step S4, an average error of a plurality of verification results obtained by K-fold cross verification is used as an index for evaluating the super-parameter combination performance.
10. The method according to claim 1, wherein in step S5, the voltage values of the proton exchange membrane fuel cell at a plurality of successive time points in the future are predicted in a plurality of steps.

Description

Proton exchange membrane fuel cell degradation trend prediction method based on frequency enhancement direct prediction Technical Field The invention belongs to the technical field of fuel cell life prediction, and particularly relates to a proton exchange membrane fuel cell degradation trend prediction method based on frequency enhancement direct prediction. Background Fuel Cells (FCs) are a key technology in the clean energy field, and the prediction of the degradation trend of long-term operation performance of the Fuel cells is of great importance for equipment health management, life assessment and maintenance decision. Currently, degradation prediction methods for proton exchange membrane fuel cells (Proton Exchange Membrane Fuel Cell, PEMFCs) are mainly classified into model driving, data driving and hybrid driving. The model driving method establishes a physical model based on an electrochemical mechanism and can explain an intrinsic mechanism of a degradation process, the data driving method directly learns a degradation rule by utilizing historical operation data through a machine learning algorithm and has stronger adaptability and flexibility, and the hybrid driving method combines the two methods and aims at considering mechanism interpretability and data fitting capability. These methods have all been applied in engineering practice and provide a useful support for status monitoring and lifetime assessment of PEMFCs. However, existing prediction methods still have several limitations. The model driving method relies on an accurate mechanism model, has difficult modeling and high calculation complexity on a degradation process under complex dynamic working conditions and multi-factor coupling, has limited capability in capturing remote dependency relationship in a long-time sequence, is easy to be interfered by noise and abnormal values, influences the stability and accuracy of prediction, and has complex structure and numerous parameters while trying to combine the advantages of the two, and often faces the problems of high calculation resource consumption, low training efficiency and insufficient generalization performance in practical application. Therefore, how to construct a method which can effectively capture the long-term nonlinear degradation characteristics of the PEMFC and has higher prediction efficiency and robustness is still a technical problem to be solved in the field. Disclosure of Invention In order to solve the technical problems, the invention provides a proton exchange membrane fuel cell degradation trend prediction method based on frequency enhancement direct prediction, so as to solve the problems in the prior art. In order to achieve the above object, the present invention provides a method for predicting degradation trend of proton exchange membrane fuel cell based on frequency enhancement direct prediction, comprising: s1, acquiring voltage time sequence aging data of a proton exchange membrane fuel cell, and preprocessing the voltage time sequence aging data to construct an input sample set; s2, constructing a ITransformer architecture-based sequence-to-sequence prediction model; S3, training the sequence-to-sequence prediction model by adopting a frequency enhancement direct prediction training paradigm, wherein the total loss function of the frequency enhancement direct prediction training paradigm is formed by weighted summation of a time domain loss component and a frequency domain loss component; S4, performing super-parameter optimization on the trained sequence-to-sequence prediction model by adopting a Bayesian optimization method, and evaluating the model performance by adopting a cross verification method to determine a final prediction model; s5, predicting the voltage degradation trend of the proton exchange membrane fuel cell by using the final prediction model. Preferably, in step S1, the preprocessing includes denoising the voltage time sequence aging data by using a variation modal decomposition method, and processing the denoised time sequence by using a sliding window method to construct the input sample set, where an input time length of each sample is greater than an output prediction time length. Preferably, in step S2, the input of the sequence to the sequence prediction model is a voltage history sequence subjected to an instance normalization process. Preferably, in step S3, the frequency domain loss component is calculated by converting the predicted sequence and the true sequence to the frequency domain by Fourier transform, respectively, and calculating the L1 norm difference of the two in the frequency domain. Preferably, in step S3, the formula of the frequency domain loss component is: ; Wherein, the For the frequency domain lost component,In order to be a true sequence of the sequences,Is a predicted sequence. Preferably, in step S3, the total loss function: ; Wherein, the For the time-domain loss component,Is a preset weight coeffici