The mixed gas concentration regression prediction method for electronic nose based on bidirectional Mamba model

Abstract Electronic noses, inspired by the mammalian olfactory system, have emerged as vital tools across various domains for gas detection and analysis. However, predicting the concentrations of mixed gases poses a significant challenge due to the nonlinear responses of gas sensors and the intricate nature of gas mixtures. To tackle this issue, this study proposes a novel bidirectional Mamba model that integrates state space modeling and bidirectional information fusion to improve the accuracy and reliability of mixed gas concentration prediction. Leveraging bidirectional time series modeling, the model effectively captures both past and future gas concentration dynamics, excelling in predicting methane, carbon monoxide, and ethylene concentrations with superior performance metrics R 2 values of 0.9989, 0.9985, and 0.9905, respectively compared to traditional methods. Its bidirectional structure and residual learning mechanism adeptly handle cross-sensitivity, long-term dependencies, and nonlinearity in gas mixtures, enabling precise global information capture from sensor responses. This work provides innovative insights and methodologies for enhancing electronic nose applications in complex gas monitoring scenarios.