Research on the Soft-sensing Modeling Method for the Naphtha Dry Point of an Atmospheric Tower

Atmospheric tower process data from an atmospheric-vacuum distillation unit present a number of uncertainties. Therefore, the soft-sensing modeling approach is affected by many factors and the universality is poor. In this paper, a soft-sensing modeling process for the dry point of an atmospheric tower overhead naphtha is proposed, and it considers various stages of soft sensing and integrates a variety of algorithms. To realize data pretreatment, wavelet and boxplot combination algorithms were used to identify anomalies based on certain variables and the Robust Partial Least Squares (RPLS) method was used to recognize multivariate anomalies. Then, the influence of the maximum delay time and the effect of the sampling interval on the model performance were investigated using the grid search algorithm. After selecting the optimal dynamic input variables, we compared the effect of the soft-sensing models using many methods and found that the Dynamic Partial Least Squares (DPLS) approach was the best modeling method. Finally, according to the characteristics of the data of the atmospheric decompression process, the use of a moving window to update the model online was proposed. The test results showed that the method had good accuracy and universality.