Answers are grounded in this paper and cite their sources.
Reliable short-horizon voltage forecasting is important for condition monitoring of proton exchange membrane fuel cell (PEMFC) systems operating under progressive degradation and changing loads. This study proposes DMC-Net, a lightweight dual-stream network that combines a selective state-space branch for slowly varying temporal evolution with a multi-scale convolutional branch for localized dynamics. Cross-attention couples the two representations before a residual voltage prediction head estimates the future interval-mean voltage. Under the unified completed-hour protocol, five-seed one-hour-ahead RMSEs are 2.173 mV on FC1, 3.232 mV on FC2, and 1.010 mV on the independent Tongji FC-DLC dataset, giving the lowest mean RMSE among the compared models in all three hourly benchmarks. Additional evaluations cover 3-h and 6-h leads, causal SG input filtering, restricted development-data availability, a 59-min cycle-level task, and same-source steady-to-dynamic transfer. At L = 60 , DMC-Net contains 0.08972 M registered parameters, with a median GPU forward latency of 1.403 ms and an estimated 0.0020650 GFLOPs under the reported computer benchmark. The resulting accuracy and compact computational footprint support DMC-Net as an efficient framework for short-horizon PEMFC voltage forecasting and condition monitoring.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: