A Novel Supercapacitor Degradation Prediction Using a 1D Convolutional Neural Network and Improved Informer Model  

作　　者：Hao Zhang Zhenxiao Yi Le Kang Yi Zhang Kai Wang 

机构地区：[1]the College of Materials Science and Engineering,Xi’an University of Science and Technol-ogy,Xi’an 710054,China [2]the School of Electrical Engineering,Qingdao University,Qingdao 266071,China [3]the School of Energy Sciences and Engineering,Nanjing Tech University,Nanjing 211816,China

出　　处：《Protection and Control of Modern Power Systems》2024年第4期51-68,共18页现代电力系统保护与控制（英文）

基　　金：funded by the Youth Fund of Shandong Province Natural Science Foundation(No.ZR2020QE212);the Key Projects of Shandong Province Natural Science Foundation(No.ZR2020KF020);the Guangdong Provincial Key Lab of Green Chemical Product Technology(No.GC 202111);the Zhejiang Province Natural Science Foundation(No.LY22E070007);the National Natural Science Foundation of China(No.52007170).

摘　　要：Safety and reliability are crucial for the next-generation supercapacitors used in energy storage systems,while accurate prediction of the degradation trajectory and remaining useful life(RUL)is essential for analyzing degradation and evaluating performance in energy storage systems.This study proposes a novel data processing and improved one-dimensional convolutional neural network(1D CNN)-informer framework for robust RUL prediction.In data preprocessing,all data from two structures are adjusted to a unified format,and cross-entropy loss is used to couple the 1D CNN and informer.Then,the minimum-maximum feature scaling method is used for normalization to accelerate the training process in reaching the minimum cost function.A relative position encoding algorithm is introduced to improve the Informer model,enabling it to better learn the sequence relationships between data and effectively reduce prediction variability.Supercapacitor data in dif-ferent working conditions are used to validate the proposed method.Compared with other existing methods,the maximum root mean square error is reduced by 32.71%,the mean absolute error is reduced by 28.50%,and R^(2) is increased by 4.79%.The strategy considers the complementarity between two single models,which can extract features and enrich local details,as well as enhance the model’s global perception ability.The experimental results demonstrate that the proposed model achieves high-precision and robust RUL prediction,thereby pro-moting the industrial application of supercapacitors.

关 键 词：SUPERCAPACITORS remaining useful life convolution neural network transformer INFORMER 

分 类 号：TM53[电气工程—电器]