A Fast Detection Method of Turbulent Gases Based on Gated Recurrent Unit and Attention Mechanism

Abstract

Unlike the stable gas mixtures often analyzed in lab settings, the gas mixtures in practice may change fast due to the turbulent conditions of the environment, making the detection of gas components challenging. There will be a great safety risk if no fast and accurate detection method. In this article, an attention-based gated recurrent unit (AGRU) is proposed to solve this problem. It was introduced in detail that the method based on the gated recurrent unit (GRU) and attention mechanism. The component and concentration of the gas mixture were analyzed simultaneously by the dual loss function. The individual gas concentration in the gas mixture was obtained by multilabel coding. It was used to evaluate the performance of the model by accuracy, root mean square error (RMSE), number of model parameters, and floating point of operations (FLOPs). The experiment result showed that the AGRU’s accuracy curve is fluctuating around 97%, and varying the length of response time between 3 and 30 s hardly affected the recognition accuracy of gas species. AGRU also has the smallest RMSE of each gas among models. At 3 s, AGRU’s FLOPs are smaller than Improved LeNet5, which lays the foundation for the deployment in the embedded. Therefore, our proposed algorithm has a higher detection performance, lower complexity, and excellent robustness to be applicable in various practical applications for fast detection.