Degradation trajectories prognosis for fuel cell based on MP-NBEATS

The performance degradation trajectory of fuel cells has strong nonlinear characteristics, and accurate and efficient long-term prediction of fuel cell performance degradation is of great significance to protect the safe operation of batteries. Since long-term forecasting of time series is difficult to predict its trend and fluctuation, this paper proposes a multi periodic neural basis expansion analysis for interpretable time series forecasting (MP-NBEATS). This method obtains multiple periods of the series by decomposing the voltage series, and integrates the prediction results of neural basis expansion analysis for interpretable time series forecasting (NBEATS) under different periods, and finally realizes long-term prediction. Compared with the traditional method, this method can better predict the trend and seasonal characteristics of the time series. Finally, through experimental verification, the error of the proposed method can reach 0.983%.