Application of AR-ELM in SOFC Parameter Identification

针对I-V特性数据集不足时会使启发式算法(MhAs)对SOFC的参数辨识精度不高的问题，本文提出了一种基于自适应正则化极限学习机算法(AR-ELM)的MhAs辨识策略——AR-ELM-MhAs。首先，引入验证误差反馈机制，根据验证集精度自适应调整传统ELM中正则化项的正则化系数，从而加强算法的泛化能力，提高数据预测精度。其次，利用改进后的AR-ELM对两种不同条件的SOFC原始I-V特性数据进行训练并预测，对数据集进行扩展，从而为MhAs的参数辨识提供更可靠的I-V特性数据。最后，将MhAs应用于扩展后的数据集的SOFC参数辨识中，对比不同数据预处理下MhAs对SOFC的参数辨识效果，以探究所提策略的可行性和改进的有效性。对比实验结果表明，AR-ELM-MhAs有效提高了SOFC参数辨识的精度。To address the accuracy degradation of metaheuristic algorithms (MhAs) in solid oxide fuel cell (SOFC) parameter identification under insufficient I-V characteristic datasets, this study proposes a novel AR-ELM-enhanced MhAs strategy (AR-ELM-MhAs). First, we introduce a validation error feedback mechanism to adaptively adjust the regularization coefficient of the regularization term in traditional Extreme Learning Machine (ELM) based on the accuracy of the validation set. This enhances the algorithm’s generalization ability and improves the accuracy of data prediction. Second, the improved AR-ELM is used to train and predict the raw I-V characteristic data of SOFC under two different conditions, thereby expanding the dataset and providing more reliable I-V characteristic data for the parameter identification of MhAs. Finally, MhAs are applied to the parameter identification of SOFC using the expanded dataset, and the parameter identification performance of MhAs under different data preprocessing methods is compared to explore the feasibility and effectiveness of the proposed strategy. The comparative experimental results indicate that the AR-ELM-MhAs approach significantly improves the accuracy of SOFC parameter identification.