(59at) Data-Driven Adaptive Sparse Identification of Time-Varying Nonlinear Dynamics for 2,3-Bdo Distillation Column

Optimizing chemical processes is a critical task that involves maximizing the production rate, minimizing energy consumption, and minimizing the environmental impact. However, achieving these goals is challenging due to the inherent complexity of chemical processes, which are often nonlinear and subject to uncertainties and disturbances. These uncertainties can arise from factors such as changes in feedstock quality, equipment degradation, and unexpected changes in operating conditions. Moreover, the time-varying nature of these disturbances and uncertainties makes it difficult to use theoretical modeling approaches to predict the behavior of the process accurately. To address these challenges, data-based modeling approaches have become increasingly popular in recent years. Data-based models can effectively capture the complex dynamics of chemical processes by using data collected from sensors, control systems, and other sources.

This study aims to develop an adaptive sparse identification model to predict the temperature of the 2,3-BDO distillation column, and its performance was evaluated. Process data were collected from a demo plant in operation at GS Caltex in Korea, and data analysis and preprocessing were conducted before building the process model. The Sparse Identification of Nonlinear Dynamics (SINDy) method proposed by Professor Brunton [1,2] was employed to construct the model, and sequentially threshold least square [3] and sequential feature selection [4] methods were used as sparse regression promotion techniques. As a case study, the performance of models was compared based on the sparse-promoting parameters of each method and the size of the training data. The development of an adaptive model is a strategy that can be employed when the performance of a deployed model deteriorates over time [5,6]. Specifically, such a model can be adapted either by retraining the model using newly measured data or by applying an update method that modifies the coefficients of an existing model. By utilizing the former approach, it was found that the temperature of the distillation column changes over time. On the other hand, the latter approach confirmed that the process could be predicted and interpreted by a single deployed model through coefficient tuning. In summary, this study provides insights into the development of an adaptive sparse identification model for predicting the temperature of the 2,3-BDO distillation column and the proposed method can be a valuable tool for process monitoring, control and optimization in non only the chemical process but also other applicable systems.

Literature cited:

1. Brunton SL, Proctor JL, Kutz JN, Bialek W. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proc Natl Acad Sci U S A. 2016;113(15):3932-3937.
2. de Silva B, Champion K, Quade M, Loiseau J-C, Kutz J, Brunton S. PySINDy: A Python package for the sparse identification of nonlinear dynamical systems from data. J Open Source Softw. 2020;5(49):2104.
3. Bhadriraju B, Bangi MSF, Narasingam A, Kwon JS Il. Operable adaptive sparse identification of systems: Application to chemical processes. AIChE J. 2020;66(11):1-15.
4. Goyal P, Benner P. Discovery of nonlinear dynamical systems using a Runge-Kutta inspired dictionary-based sparse regression approach. Proc R Soc A Math Phys Eng Sci. 2022;478(2262).
5. Sun Z, Tang J, Qiao J, Cui C. Review of Concept Drift Detection Method for Industrial Process Modeling. Chinese Control Conf CCC. 2020;2020-July:5754-5759.
6. Bang SH, Ak R, Narayanan A, Lee YT, Cho H. A survey on knowledge transfer for manufacturing data analytics. Comput Ind. 2019;104(November 2018):116-130.