Rapid Gas Sensing Based on Pulse Heating and Deep Learning

We have reported a rapid sensing method for MEMS gas sensors based on the combination of pulse heating and deep learning. Pulse heating is used to reduce the power consumption of MEMS gas sensor and to realize the rapid gas sensing process. The output signals of the gas sensor are extracted and assigned with different weights according to their importance by deep learning algorithm during data processing. Afterwards, we have used two prediction methods to verify the accuracy of this method. One is to randomly select part of the measurement data as the test sets, which results in an average relative error of 4%. The other method is using the data of the first two days as the training sets to build the model, which could be used to predict the test sets of the third day and the best average relative error of 3.2% is achieved.