Accurate identification of mixed gases under dynamically varying concentration conditions is a critical challenge in environmental monitoring and industrial safety. However, the performance of existing methods is severely constrained by sensor cross-sensitivity and weak transient responses. To address these limitations, this paper proposes a differentially enhanced parallel rotating neuron reservoir computing model. The proposed model incorporates dual-path multi-scale differential mechanisms at both the input and output stages, thereby enhancing the sensitivity to dynamic concentration variations from the raw signal domain and the high-dimensional state space, respectively. In addition, a parallel rotating neuron reservoir architecture is employed to efficiently process multi-channel signals and to improve feature diversity. Experiments conducted on the UCI dynamic mixed-gas dataset demonstrate that the proposed model achieves an accuracy of 97.8% in a four-class methane/ethylene classification task and attains normalized mean squared errors of 0.0465 and 0.1359 for methane and ethylene, respectively. These results indicate that the proposed approach provides an effective, training-efficient edge intelligence solution for enhancing the detection efficiency of electronic nose systems.
Zhou et al. (Mon,) studied this question.