Efficient degradation prediction of PEMFCs using ELM-AE based on fuzzy extension broad learning system

Abstract

Degradation prediction is an important part of proton exchange membrane fuel cells (PEMFCs) prognostic research, and it is essential for prolonging the durability of PEMFCs. An accuracy and efficient degradation prediction of PEMFCs method is proposed, which combines extreme learning machine based on auto-encoder (ELM-AE) and fuzzy extension broad learning system (FEBLS). The ELM-AE inherits the fast learning ability of ELM while retaining the dimensionality reduction and feature learning ability of AE. Specifically, the proposed FEBLS consists of three parts, the Takagi–Sugeno–Kang (TSK) fuzzy subsystems, enhancement layer, and extension layer. First, a set of TSK fuzzy subsystems are utilized to replace the mapping nodes of the existing broad learning system (BLS), where the latent representation of the power time series of PEMFCs are processed. Then, the outputs of all fuzzy subsystems are sent to the next layer for nonlinear transformation to obtain enhancement nodes. After that, the enhancement nodes are further fed into the latest layer for further nonlinear transformation to obtain the extension nodes. Finally, the accuracy and efficiency of the proposed ELM-AE-FEBLS method are verified by two different aging datasets of PEMFCs. Compared with ELM, ELM based on particle swarm optimization, and BLS methods, the comparison analysis shows that ELM-AE-FEBLS of multi-input variables with power time series has more competitive in accuracy. Therefore, the proposed method can accurately predict the remaining useful life of fuel cells.