This paper presents a data-driven short-term voltage prediction model for
high-power fuel cells under three operating conditions: idle, rated, and
variable load. Long short-term memory (LSTM) recurrent neural network has good
performance in voltage prediction, but the accuracy of prediction for volatile
voltage data is significantly reduced. In this paper, the obvious fluctuations
caused by voltage recovery due to resting are discussed, and the proposed model
is optimized for this phenomenon from two perspectives: the method I is the
wavelet algorithm is used to extract features from the raw voltage data, and
then the decomposed waveform is predicted using LSTM, lastly the predicted
results are synthesized into the final voltage trend using inverse wavelet; the
method II is to divide the voltage loss into reversible loss and irreversible
loss parts, the reversible loss part is predicted by exponential model, while
the irreversible loss part by LSTM, then the final results are obtained by
superimposing the two parts. The results show that method I has better
optimization effect under variable load and rated conditions, and method II has
higher accuracy for idle condition. With the optimized model to predict the
voltage more accurately, it can help to adjust the unsuitable operation
condition timely, making the voltage decline in slower speed, that improves the
durability of the fuel cells.