A Concentration Prediction and Gas Classification Model Based on LSTM-Attention Multi-task Learning Framework Network

Abstract

The electronic nose (E-nose), a bionic olfactory system, has been widely used in gas identification and concentration prediction. However, these tasks are usually based on separate systems, leading to high detection costs. To address this issue, a multi-task learning network model based on LSTM-Attention (MTL-LSTMA) as a skeleton has been proposed to simultaneously train both species recognition and concentration prediction tasks. The introduction of an attention mechanism greatly reduces interference in gas feature information, improving the electronic nose pattern recognition algorithm’s efficiency. The MTL-LSTMA model was tested through five sets of comparison experiments, showing the best performance in both concentration prediction and species identification. The proposed model has broad application prospects and may revolutionize gas detection technology.