Development of an ANFIS-PSO-Based Digital Twin and Model Predictive Control Framework for Intelligent Valve Regulation in Supercritical CO₂ Extraction of Silkworm Pupae Oil

Supercritical carbon dioxide (SC-CO₂) extraction has emerged as an environmentally friendly and highly efficient technology for recovering high-value bioactive compounds from silkworm pupae. However, maintaining stable extraction conditions through accurate valve regulation remains a challenging control problem because of the nonlinear dynamics and complex interactions among process variables. This study proposes an intelligent control framework that integrates an Adaptive Neuro-Fuzzy Inference System optimized by Particle Swarm Optimization (ANFIS-PSO), a data-driven digital twin, and Model Predictive Control (MPC) to improve valve regulation during the SC-CO₂ extraction process.

A comprehensive experimental dataset consisting of 300 operating samples was established using laboratory-scale extraction experiments conducted under various operating conditions. The dataset includes extraction temperature, extraction pressure, processing time, and valve opening conditions covering the practical operating range of the extraction system. Experimental measurements of extraction yield were collected and subsequently employed to develop a high-fidelity data-driven model representing the dynamic behavior of the extraction process. The developed ANFIS-PSO model was trained using the experimental data to accurately describe the nonlinear relationship between process variables and valve opening requirements. Compared with the conventional ANFIS and ANFIS-GA models, the proposed ANFIS-PSO model demonstrated the highest prediction accuracy, achieving RMSE = 3.53, MAE = 2.76, and R² = 0.934, and was therefore selected as the internal prediction model of the digital twin.

The developed digital twin was subsequently integrated with a Model Predictive Control strategy to realize intelligent closed-loop valve regulation. The performance of the proposed ANFIS-PSO-MPC controller was evaluated against a conventional PID controller under identical operating conditions. Simulation results demonstrated that the proposed controller significantly improved dynamic control performance by reducing overshoot from 1.286% to 0.021%, shortening the rise time from 107 min to 92 min, decreasing the settling time from 166 min to 139 min, and practically eliminating steady-state error from 0.90% to 0.002%. These improvements were achieved while maintaining stable control actions and accurate.