An Algorithm Based on Transformer and Temporal Convolution Network for Early Identification of Ternary Gas Mixture

Metal oxide (MOX) gas sensor arrays play an important role in various fields of gas detection, but their development is also limited by their performance deficiencies, such as measurement delays due to slow response times, and cross sensitivity interfering with gas identification. In addition, gas identification typically requires complete time series data of the steady-state response and reaction of the sensor array, which affects the efficiency. In this article, we propose a novel algorithm transformer equipped temporal convolution network (TTCN) based on the transformer and temporal convolution network (TCN) structure that can automatically perform feature extraction and gas mixture recognition on time series data before reaching equilibrium, overcoming the recognition difficulties caused by measurement delays and measurement interferences. This algorithm extracts global and local features using the attention mechanism in the transformer structure and multiscale convolution in the TCN structure to acquire instantaneous information on changes in the trends of gases for improved gas identification. The TTCN provides precise identification of early gas data and identifies ternary mixtures of formaldehyde, ethanol, and acetone with an average identification accuracy of 98.23%. In this study, we carry out in-depth tests to confirm the efficacy of our proposed algorithm and to show its significant advantages over other algorithms. Importantly, the excellent identification performance of the TTCN in the early stages of gas exposure demonstrates its significance for future real-time applications.