AI enhanced model predictive control for optimizing LPG recovery through integrated computational modeling design of...

Area of Science:

• Chemical Engineering
• Process Control
• Artificial Intelligence

Background:

• Debutanizer columns face challenges in optimizing Liquefied Petroleum Gas (LPG) recovery due to operational complexities.
• Conventional control methods often fall short in maintaining peak efficiency and stability under dynamic conditions.

Purpose of the Study:

• To develop and evaluate an integrated framework combining Response Surface Methodology (RSM) with Model Predictive Control (MPC) for enhanced LPG recovery.
• To compare the performance of conventional MPC with Artificial Intelligence-enhanced MPC (AI-enhanced MPC) in a debutanizer column simulation.

Main Methods:

• Steady-state and dynamic simulations were performed using Aspen HYSYS.
• Response Surface Methodology (RSM) analyzed key process variables from steady-state simulations.
• A regression decision tree model was developed in MATLAB to support AI-enhanced MPC implementation.

Main Results:

• MPC improved LPG recovery from 99.73% to 99.85% and reduced energy consumption.
• AI-enhanced MPC further increased recovery to 99.9% and achieved greater reductions in reboiler and condenser duty.
• Both control systems effectively managed the impact of feed molar flow, enhancing thermal stability and product quality.

Conclusions:

• The hybrid framework integrating RSM with MPC and AI-enhanced MPC offers a novel approach for optimizing LPG recovery.
• AI-enhanced MPC demonstrates superior performance in improving energy efficiency, product quality, and dynamic stability in debutanizer operations.
• The proposed methodology provides a robust solution for industrial LPG recovery enhancement.