State of Health Estimation and Prediction of Fuel Cell Stacks in Backup Power Systems

Abstract

Fuel cell based backup systems are used as telecommunication power supplies, where availability of power is crucial for a reliable service. State of health (SOH) is a useful metric in aiding predictive maintenance actions, that aim to reduce the system down time as well as the operating cost. In this paper, voltage, current, and temperature data from numerous stacks installed in the field to estimate an SOH metric. A long short term memory (LSTM) recurrent neural network (RNN) is trained to predict SOH values six months into the future. Finally, the RNN performance is evaluated on prediction horizons of six months, as well as longer horizons of twelve, eighteen, and twenty-four months.