Research on binary gas mixtures identification method: Humidity interference suppression using linear discriminant analysis

When utilizing metal oxide semiconductor (MOS) sensors for gas mixtures detection, challenges such as low detection accuracy and susceptibility to interference from ambient humidity often arise. This research presents a method for effectively identifying gas mixtures in the presence of humidity interference. Using a single commercial sensor MQ3, dynamic response data for methanol-ethanol mixture was collected across various relative humidity levels through dynamic temperature modulation. Dynamic temperature modulation allows the sensor response to be extended in the time domain to include both gas concentration and humidity information. The linear discriminant analysis (LDA) method was used to suppress the humidity interference in the response, and the gas concentration information was retained to the greatest extent. Subsequently, the data was transformed into a two-dimensional matrix using the gramian angular field (GAF) algorithm, which was then fed into the self-developed multi-head feature extraction convolutional neural network (MFE-CNN) for concentration prediction. Experimental results demonstrate that this approach effectively reduces the impact of humidity interference, outperforming other control strategies. This method is anticipated to serve as a novel strategy for the detection of gas mixtures in the presence of humidity interference.