Hydrogen Leakage Detection in Confined Spaces with Drift Suppression Based on Subspace Alignment

Gas sensor has been widely used in flammable gas detection. In order to solve the problem that gas sensors are susceptible to drift in hydrogen leakage detection in confined spaces, a drift compensation framework based on subspace alignment is proposed. First, global domains and subdomains are aligned simultaneously in the new subspace, thus reducing the distribution difference between the source domain and the target domain. Then, the proposed method utilizes extreme learning machine (ELM) to iteratively refine the prediction results of the target domain, and continuously optimizes the subspace and classifier. In this way, the proposed method realizes drift compensation at the feature level. Compared with the existing methods, the proposed method achieves the highest accuracy of 79.83% in the long-term drift scenario. Therefore, the experimental results show that the proposed method is competent for hydrogen leakage detection with drift, and can provide a reference for the design of drift compensation method based on gas sensors.