改进DBO优化CRJ网络的PEMFC剩余使用寿命预测  

Improved DBO for optimizing CRJ network in predicting remaining useful life of PEMFC

作　　者：王基臣 许亮[1] 张紫叶 WANG Jichen;XU Liang;ZHANG Ziye(Tianjin Key Laboratory of New Energy Power Conversion,Transmission and Intelligent Control,School of Electrical Engineering and Automation,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学电气工程与自动化学院,天津市新能源电力变换传输与智能控制重点实验室,天津300384

出　　处：《电源技术》2024年第11期2295-2303,共9页Chinese Journal of Power Sources

基　　金：国家自然科学基金项目(61975151,61308120)。

摘　　要：质子交换膜燃料电池(PEMFC)剩余使用寿命(RUL)预测是其健康管理和故障诊断的主要问题。针对此问题提出一种基于改进蜣螂算法(HDBO)优化确定性循环跳跃储备池网络(CRJ)的PEMFC剩余使用寿命预测方法。采用局部加权回归散点平滑法对PEMFC运行数据进行重构和平滑处理,有效地滤除噪声并保存了原始数据的主要特征。利用最大信息系数(MIC)结合贝叶斯信息准则(BIC)方法选择出8个最优输入特征。采用混沌映射初始化种群、自适应调整搜索策略、引入柯西变异改进了蜣螂优化算法,利用HDBO优化CRJ的三个关键参数,建立高效的预测模型。将最优特征集作为预测模型的输入实现PEMFC的剩余使用寿命预测,实验结果表明,该方法的决定系数(R2)、平均绝对值误差(MAE)和均方根误差(RMSE)分别为0.95022、0.0025729和0.0035232,与DBO、SSA和CRJ相比,该方法的预测精度更高。Predicting the remaining useful life(RUL)of proton exchange membrane fuel cell(PEMFC)is a key challenge in their health management and fault diagnosis.A PEMFC RUL prediction method based on an improved dung beetle optimizer(HDBO)optimized deterministic recurrent jump reservoir pool network(CRJ)was proposed.The PEMFC operational data was reconstructed and smoothed using the Local Weighted Regression Scattered Point Smoothing method,effectively eliminating noise while preserving the main features of the original data.Eight optimal input features were selected using the Maximum Information Coefficient(MIC)combined with the Bayesian information criterion(BIC)method.The dung beetle optimizer was enhanced by chaos mapping for population initialization,adaptive adjustment of search strategy,and the introduction of Cauchy mutation.The HDBO optimized three key parameters of CRJ to establish an efficient prediction model.The optimal feature set was used as the input for the prediction model to achieve PEMFC RUL prediction.Experimental results demonstrate that the proposed method achieves higher prediction accuracy with R2,MAE,and RMSE values of 0.95022,0.0025729,and 0.0035232,respectively.Compared to differential evolution,simplex search algorithm,and CRJ,the proposed method exhibits superior predictive performance.

关 键 词：质子交换膜燃料电池 回声状态网络 蜣螂优化算法 剩余寿命 

分 类 号：TM911[电气工程—电力电子与电力传动]