基于BO-BiLSTM的超级电容器剩余寿命预测  

Remaining useful life prediction of supercapacitor based on BO-BiLSTM

作　　者：沈伟豪 林文文[1,2] 楼功茂 SHEN Weihao;LIN Wenwen;LOU Gongmao(School of Mechanical Engineering and Mechanics,Ningbo University,Ningbo 315211,China;Institute of Advanced Energy Storage Technology and Equipment,Ningbo University,Ningbo 315211,China)

机构地区：[1]宁波大学机械工程与力学学院,浙江宁波315211 [2]宁波大学先进储能技术与装备研究院,浙江宁波315211

出　　处：《电工电能新技术》2023年第4期59-67,共9页Advanced Technology of Electrical Engineering and Energy

基　　金：国家自然科学基金项目(22078164)。

摘　　要：为了提高超级电容器剩余使用寿命的预测精度,本文提出了一种贝叶斯优化与双向长短时记忆神经网络结合的预测模型(BO-BiLSTM),利用长滑动窗口处理容量数据来提高模型对容量衰退趋势的学习能力,达到对超级电容器剩余寿命精确预测的目的。通过对输入特征的研究和对比,选定了容量和循环数作为模型的输入,随后对滑窗大小、模型步长进行研究,发现长滑窗是模型成功的关键因素。实验模型的精度可以达到AEP=1.02%、RMSE=2.57%。在使用贝叶斯优化算法优化模型参数后,最终预测精度可以达到AEP=0.59%、RMSE=2.16%,具有较高的预测精度。In order to improve the prediction accuracy of the remaining service life of supercapacitors,a prediction model combining Bayesian optimization and bidirectional long-short-term memory neural network was proposed(BO-BiLSTM),which used the long sliding window to process the historical data to improve the model’s learning of the decline trend,and accurately predicted the remaining useful life of supercapacitors.Firstly,through the study and comparison of input characteristics,the capacity and number of cycles were selected as the inputs of the model.Then the sliding window size and model step size were studied,and it was found that the long sliding window would become the key factor for the success of the model.The accuracy of the experimental model can reach AEP=1.02%and RMSE=2.57%.After using Bayesian optimization algorithm to optimize the model parameters,the final prediction accuracy can reach AEP=0.59%and RMSE=2.16%,which has a high prediction accuracy.Finally,the robustness of the model is verified by making predictions for other supercapacitors.

关 键 词：超级电容器 剩余使用寿命 长滑窗 贝叶斯优化 双向长短时记忆神经网络 

分 类 号：TM53[电气工程—电器] TP183[自动化与计算机技术—控制理论与控制工程]