A multi-task learning framework for gas detection and concentration estimation

Abstract

Advancements in sensor devices have resulted in increased use of gas detection and concentration estimation in various fields. An electronic nose (E-nose) is an intelligent sensing device that mimics the human olfactory system and has been applied in gas analysis. We propose a multi-task learning–long short-term memory (MLSTM) recurrent network for simultaneous gas detection and concentration estimation, in which the two tasks share underlying features. The network utilizes the synergy between both two tasks, thereby enhancing the individual performances. Particle swarm optimization (PSO) proved a good method for optimizing the hyper-parameters used in the framework. The effectiveness of the framework and the parameter optimization method were verified by comparing the performances of the MLSTM model with those of eight models (two multi-task models, three classification models, and three regression models) using three datasets (two publicly available datasets and one dataset from the E-nose). The results showed that the proposed MLSTM model exhibited the best performance for simultaneous gas detection and concentration estimation because it was able to capture both global and local information.