Intelligent Multigas Monitoring: A Reconfigurable RFID Sensor with Machine Learning-Assisted Decoding for O₂ and CO₂

This paper proposes and designs a reconfigurable antenna-based multifrequency encodable RFID sensor integrated with a random forest machine learning algorithm, enabling full-spectrum, dual-parameter, high-precision synchronous wireless monitoring of O₂ and CO₂ in fermentation tank environments. First, a reconfigurable RFID tag antenna based on two sets of complementary split-ring resonators (CSRR) and photodiodes is designed, with each set integrating one sensing ring and four encoding rings to form 28 coding combinations. This design significantly improves the system's coding capacity and multimeasurement signal resolution, and it is the first application of reconfigurable optically controlled antenna technology to wine fermentation gas monitoring. Second, SnS₂/ZnO/NiO and SnO₂/CuO/TiO₂ nanocomposites are employed as gas-sensitive layers. The interfacial synergistic effect enhances gas adsorption and conductivity modulation, realizing direct wireless conversion from chemical gas signals to radio-frequency (RF) signals and addressing the limitations of narrow detection range and low sensitivity in traditional materials. Finally, a multidimensional RF feature-decoding framework based on random forest regression is constructed. By extracting multidimensional features such as resonant frequency shift and amplitude variation, a nonlinear mapping model from RF responses to gas concentrations is established, breaking the accuracy bottleneck of traditional linear fitting methods in multidimensional nonlinear signal inversion. Experimental results show that the sensor can detect O₂ within the range of 1000-250,000 ppm, with a response/recovery time of 29.8/39.4 s, an amplitude change of 14.79 dB, and a linear fitting R² of 0.99523; for CO₂, the detection range is 500-50,000 ppm, with a response/recovery time of 27.3/35.8 s, an amplitude change of 17.13 dB, and a linear fitting R² of 0.99894, demonstrating excellent repeatability and long-term stability. The random forest model shows high accuracy and strong generalization ability in gas concentration inversion, and its prediction results are highly consistent with the actual values, significantly outperforming traditional fitting methods. This research facilitates the digital and refined transformation of the traditional brewing industry. It not only improves the stability of product quality and reduces energy consumption and labor costs but also provides core support for the upgradation of industrial technical standards and the collaborative innovation among industry, universities, and research institutions.