Abstract
The proton exchange membrane fuel cell (PEMFC) is an extremely clean and efficient power generation device. However, its limited lifespan has restricted the large-scale commercial development of PEMFCs. Life prediction is a promising solution for the further life extension of PEMFCs. In this paper, D-S ELM(DWT-SaDE ELM), define as, an enhanced extreme learning machine (ELM) optimized by discrete wavelet transform (DWT) and self-adaptive differential evolutionary algorithm (SaDE), is proposed to predict the remaining useful life (RUL) of PEMFCs. In D-S ELM, DWT is employed to extract available features from multi-input data with stochastic noise. Then, SaDE explores the optimal parameter configuration for the ELM neural network. Moreover, the influence of training data sizes on the prediction results is discussed. Simulations show that D-S ELM has obvious advantages in prediction accuracy. Furthermore, the superiority of D-S ELM in small sample applicability, prediction speed and robustness make it more suitable for the online prediction of PEMFCs.
Highlights
Energy issue is the hottest topic in the 21th century
least square support vector machine (LSSVM) has good prediction performance in processing large sample data and regularized particle filter (RPF) improves the accuracy by solving the problem of particle diversity loss
This paper proposes D-S ELM, which improves the prediction accuracy of extreme learning machine (ELM) while retaining the superiority of speed to predict the remaining useful life (RUL) of proton exchange membrane fuel cell (PEMFC), which lays a foundation for its online real-time prediction
Summary
Energy issue is the hottest topic in the 21th century. Hydrogen energy is attracting more and more attention due to its high efficiency, cleanliness and sustainability. Mona Ibrahim et al [9] proposed a data driven method based on an auto-regressive integrated moving average model (ARIMA) using discrete wavelet transform (DWT). Rui Ma et al [14] presented a grid long short-term memory (G-LSTM) neural network which can achieve an accurate voltage degradation curve when solving complex prediction problems. LSSVM has good prediction performance in processing large sample data and RPF improves the accuracy by solving the problem of particle diversity loss. Previous data-driven algorithms (e.g., ESN and LSTM etc.) focus on improving the prediction accuracy. These algorithms with complex network structures and long prediction time cannot be applied on-line. This paper proposes D-S ELM, which improves the prediction accuracy of extreme learning machine (ELM) while retaining the superiority of speed to predict the RUL of PEMFCs, which lays a foundation for its online real-time prediction
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
