Gas Sensor Drift Compensation Using Multi-model Ensemble Networks

In gas sensing, a sensor drift is caused by the gradual chemical deterioration in sensor responses over an extended period. It is an inevitable and challenging issue and affects the detection performance of the sensor. A recently developed network model with a deep-learning scheme is considered as a promising solution to overcome such gas sensor drift problems. In this context, we present an improved multi-model ensemble network consisting of tabular data learning network (TabNet), support vector machine (SVM), and long short-term memory (LSTM). The proposed TabNet–SVM–LSTM (TSL) ensemble was implemented by employing a software-based approach for the gas sensor drift compensation without modifying hardware components. To validate the proposed TSL ensemble, a gas sensor drift dataset obtained from the UCI machine learning repository was considered. The dataset spans three years and comprises data from sixteen metal-oxide gas sensors monitoring six distinct volatile organic compounds under tightly controlled operational conditions. The extensive experiments showed that the TSL ensemble can adeptly manage sensor drift and outperform baseline competitive schemes for gas sensor drift compensation.