A Novel Calibration Scheme of Gas Sensor Array for\na More Accurate Measurement Model of Mixed Gases

Gas\nsensor arrays (GSAs) usually encounter challenges due to the\ncross-contamination of mixed gases, leading to reduced accuracy in\nmeasuring gas mixtures. However, with the advent of artificial intelligence,\nthere is a promising avenue for addressing this issue effectively.\nIn pursuit of more accurate mixed gas measurements, we proposed a\nmeasurement model leveraging neural networks. Our approach involved\nemploying the encoder of an autoencoder network (AEN) to extract features\nfrom experimental data, while fully connected layers were utilized\nfor predicting concentrations of mixed gases. To refine the neural\nnetwork parameters, we employed a variational autoencoder to generate\nadditional data resembling the distribution of experimental data.\nSubsequently, we designed a domain difference maximum entropy technique\nto identify optimal concentration points for the calibration data.\nThese calibration points were instrumental in training the fully connected\nlayers, enhancing the model’s accuracy. During practical usage,\nwith the AEN configuration fixed, the model can be fine-tuned by using\na small subset of test points across large-scale GSA deployments.\nSimulation and practical measurement results demonstrated the efficacy\nof our proposed measurement model, boasting high accuracy, with confidence\nintervals for relative errors of the four gas measurements below 3%\nat the 95% confidence level. Besides, the calibration scheme reduced\nthe number of test points compared with traditional methods, reducing\nthe cost of labor and equipment.