Recursive performance prediction of automotive fuel cell based on conditional time series forecasting with convolutional neural network

Abstract

The high efficiency and environmental friendliness make hydrogen and fuel cell technology important players in today’s energy market. However, the durability issue is a major setback for fuel cells’ adoption in various applications, including automotive. As automotive fuel cells exhibit non-linearity and dynamic behavior, predicting their performance has become a complex task since their response changes rapidly to varying operating conditions. To enhance performance prediction reliability, this study proposes a novel recursive multi-step performance prediction method for automotive fuel cells based on conditional time series forecasting with convolutional neural network (CNN). The method incorporates a conditional CNN architecture to extract representative features from the input data and a recursive strategy to iteratively predict future values based on previous predictions. Moreover, the seasonality in the data is detected to correct the prediction results by considering the intermittent performance recoveries. Experimental results show that the proposed approach achieves high prediction accuracy, particularly for long-term performance prediction. The proposed method could have significant implications for improving the reliability and efficiency of automotive fuel cell systems.