Dynamic Subspace Models for High-Sulfur Gas Sweetening Process Monitoring

Abstract

The acid component in high-sulfur gas (HSG) is as high as 5%-15%, which will causes great security risk and has strong corrosive effect to the sweetening equipment. Therefore, productive process monitoring has important significance to ensure the system normal work and safety. However, due to the time lag shift of the operating parameters, static methods seem powerless. Thus, in this paper, we proposed a dynamic subspace model for HSG sweetening process monitoring. Specifically, we first introduce the operating parameters and time series as input candidate features, and get the dynamic expansion matrix by time-lag order analysis. Then we combined the matrix with static PCA model to achieve the process monitoring. The experiments on the actual data of a HSG sweetening plant show that, the false alarm rate (FAR) and missing alarm rate (MAR) of Hotelling T2 and Q statistic (SPE) in DPCA are much lower than those of PCA. In particular, compared to PCA, the SPE MAR in DPCA reduced by 13.42%, and the T2 MAR in DPCA is 4 times lower than that of PCA. Therefore, the proposed monitoring method for HSG sweetening process is efficient and feasible.