基于集成型极限学习机的氢燃料电池寿命预测  

Research on PEMFC Lifetime Prediction Based on Ensemble Extreme Learning Machine

作　　者：杨淇 陈景文[1] 华志广[2] 李祥隆 赵冬冬[2] 兰天一 窦满峰[2] Yang Qi;Chen Jingwen;Hua Zhiguang;Li Xianglong;Zhao Dongdong;Lan Tianyi;Dou Manfeng(School of Electrical and Control Engineering Shaanxi University of Science and Technology,Xi’an 710021 China;School of Automation Northwestern Polytechnical University,Xi’an 710021 China;School of Materials Science and Engineering Beijing University of Chemical Technology,Beijing 100029 China)

机构地区：[1]陕西科技大学电气与控制工程学院,西安710021 [2]西北工业大学自动化学院,西安710021 [3]北京化工大学材料科学与工程学院,北京100029

出　　处：《电工技术学报》2025年第3期964-974,共11页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金(52307251,52277226);中国博士后科学基金(2023TQ0277,2023M742840)资助项目。

摘　　要：基于数据驱动的寿命预测方法能精准预测质子交换膜燃料电池(PEMFC)的剩余使用寿命,提高预测性能是当前寿命预测领域的研究热点。针对PEMFC寿命预测过程中预测精度与鲁棒性的提升问题,基于统计学原理的寿命预测方法,提出一种集成极限学习机(EELM)结构,对PEMFC的寿命进行长期预测。集成结构中包含了50次重复测试,通过局部强化优化器算法对每次测试结果进行优化,提升了寿命预测精度。在长期预测的结果中,给出了EELM预测结果的平均值和95%置信区间,提升了系统的鲁棒性。最后采用稳态电流、准动态电流条件和动态电流下的老化数据集验证了所提方法的有效性与可行性。Data-driven methods can accurately predict the remaining useful lifetime of the proton exchange membrane fuel cell(PEMFC),the improvement of the prediction performance is the current research hotspot in the field of lifetime prediction.Aiming to improve prediction accuracy and robustness in the lifetime prediction field of PEMFC,an ensemble extreme learning machine(ELM)structure is proposed to predict the lifetime of PEMFC in the long term based on the statistical lifetime prediction principle.The ensemble structure contains 50 times repetitive tests,and for each ELM,it is optimized by the partial reinforcement optimizer algorithm,which improves the lifetime prediction accuracy.Firstly,the aging data were filtered by the moving average filtering method to filter out the noise and spikes to get smooth data.Then an ensemble extreme learning machine(EELM)model is introduced to predict the lifetime of the PEMFC in the long term by independent ELMs in the ensemble structure.The EELM model adopts a multiinput structure and optimizes the input weights and hidden layer bias by Partial Reinforcement Optimizer to improve the model's generalization ability and prediction accuracy.After that,the prediction results are assembled,and the ensemble structure contains 50 independent ELMs,and the 50 times prediction results are statistically analyzed.Assuming the weights of the 50 times predictions are the same,the final prediction value is obtained by averaging the predictions at each time point.In the results of the long-term prediction,the average and 95%confidence interval of the prediction results of the ensemble ELM are given.To verify the validity and feasibility of the proposed method,three sets of aging data sets under steady-state,quasi-dynamic and dynamic conditions are verified.Based on the experiments,it can be obtained that the RMSE of the proposed method is 0.01044 and the MAPE is 0.2218%under steady state,0.02373 and 0.6296%under quasi-dynamic,and 0.004861 and 7.716%under dynamic.Comparing the prediction results wi

关 键 词：质子交换膜燃料电池 极限学习机 集成结构 局部强化优化器 

分 类 号：TM911.42[电气工程—电力电子与电力传动]