Gas Component Strategy and Signal Graph Attention Network for Enhanced Gas Concentration Estimation

With advancements in intelligent sensing technology, artificial olfactory systems are increasingly essential in various applications. However, conventional concentration estimation methods face notable challenges, such as intensive training demands, high memory consumption, and limited scalability, all of which restrict their industrial applicability. To overcome these limitations, we propose a Gas Component Strategy (GCS) that enhances scalability and reduces training complexity by estimating concentrations via dedicated models for individual gas components. Building on this strategy, we introduce the Signal Graph Attention Network (SGAN), a Graph Neural Network-based approach that models dynamic interactions among sensors. SGAN captures intricate spatial and temporal dependencies, enabling more accurate gas concentration estimations. Experimental evaluations on a custom dataset demonstrate that SGAN outperforms existing models, achieving up to 35% lower MAE and RMSE, with R² improvements of approximately 5%. GCS and SGAN offer a scalable and accurate solution for precise gas concentration estimation, advancing the applicability of artificial olfactory systems in industrial settings.