Intelligent Soft Sensor Design for Composition Prediction in a Debutanizer Column Using ANN

Abstract

Chemical industries, such as natural gas and refinery industrial processes, face delays in measuring the composition of butane, as normal measurements depend on lab sampling. This delay affects the operation of the debutanizer to maintain the product quality. The single dynamic neural network model is used to overcome the lab-measurement delay through the development of a data-driven soft sensor, which enables the monitoring of the product quality at the top and the bottom of the column. The process of the debutanizer is simulated using Aspen-HYSYS in both steady-state mode and dynamic mode. The model is validated with life data from a natural gas plant, which is then used to develop the neural network model. The principal component analysis is performed using RStudio as a data reduction tool to optimize the number of variables used to create the soft-sensor model. The MATLAB neural network time series toolbox is used to build the neural network model and train it to give the future measurement of isobutane and normal butane. The function of the network is deployed using HYSYS-MATLAB interface code. According to the results from the test data, the built soft sensor showed accurate measurements of the product quality. The developed ANN-based soft sensor demonstrated exceptional accuracy, achieving a regression coefficient of 0.9999 for all data and 0.9987 for validation data, with a mean squared error of 1.07 and an RMSD of 1.88. The prediction accuracy exceeded 99%, and the minimal autocorrelation of error within the 95% confidence limit confirmed the robustness and reliability of the proposed model.