Methane Concentration Prediction of Hydrogen Production Process Based on Stacked Regularization Quality-driven Autoencoder

Hydrocarbon steam reforming is a widely used hydrogen production method in refineries. In this process, the concentration of methane (CH4) at the outlet of the reformer furnace is the critical quality indicator. Therefore, it is necessary to predict it accurately. However, the mechanism of the reformer reaction process is complex in practice. Data-driven soft sensor technology is an effective way to realize the online prediction of the concentration of CH 4. Considering the characteristics of strong nonlinearity, a stacked regularization quality-driven autoencoder (SQAER) is proposed in this paper. The model constructs a deep SQAER model by extracting the quality-related feature vectors layer by layer, mining the intrinsic relationship between CH4 concentration and auxiliary variables, relieving overfitting, and predicting the CH4 concentration accurately. The validity and effectiveness of the proposed method are verified by the hydrogen production process. The experimental results show that the proposed SQAER method can have higher accuracy and better generalization ability in the CH 4 prediction at the outlet of the reformer furnace.