A multi-algorithm fusion model for predicting automotive fuel cell system demand power

The dynamic of a full-power fuel cell vehicle highly depends on the proton exchange membrane fuel cell power response. However, when frequent load changes in full-power fuel cell vehicle, proton exchange membrane fuel cell air shortage often occurs, leading to its poor power response. Predictive control of proton exchange membrane fuel cell is considered an effective solution, in which its demand power is a key input variable and must be predicted accurately. Proton exchange membrane fuel cell demand power has a strong nonlinear trend, which leads to large prediction errors. This paper proposes a hybrid prediction model that further extracts the data trendiness through the seasonal and trend decomposition using loess method and extracts the global features using the attention mechanism. The proposed prediction model is validated using real data under different driving conditions. Experimental results show that the proposed prediction model captures data characteristics and trendiness well and can better predict proton exchange membrane fuel cell demand power compared to other models. The proposed prediction model provides a prerequisite for improving the proton exchange membrane fuel cell power response and dynamics of full-power fuel cell vehicle. [Display omitted] •A hybrid prediction model for PEMFC demand power is proposed.•Prediction length is determined for different driving conditions.•The STL method is used for efficiently extracting data series trendiness.•Proposed prediction model has better performance than other two models.•PEMFC demand power prediction favors subsequent power response control.