Improved ANFIS combined with PID for extractive distillation process control of benzene–isopropanol–water mixtures

• Published in: Chemical engineering science 2023-04, Vol.269, p.118464, Article 118464
• Main Authors: Shan, Baoming, Pang, Yanshuo, Zheng, Qi, Xu, Qilei, Wang, Yinglong, Zhu, Zhaoyou, Zhang, Fangkun
• Format: Article
• Language: English
• Subjects: Adaptive neuro-fuzzy inference system; Dynamic control; Extractive distillation; Genetic algorithm; Optimisation; Particle swarm optimisation
• ISSN: 0009-2509; 1873-4405
• DOI: 10.1016/j.ces.2023.118464

[Display omitted] •A novel adaptive neuro-fuzzy inference system and PID (ANFIS-PID) fusion control scheme was proposed for product composition control in extractive distillation.•ANFIS-PID fusion control was optimised and improved using the genetic algorithm (GA) and the particle swarm optimisation (PSO) algorithm separately.•The number of clusters (NC) as a key indicator was analysed to improve the prediction and control ability of ANFIS.•The proposed PSO-optimised ANFIS-PID fusion control scheme exhibited superior dynamic performance under ±20% feed disturbances. Real-time measurement and control of product composition in extractive distillation continues to remain a challenge; however, composition control has demonstrated the ability to stabilise control systems. In this study, a dynamic control scheme is proposed that uses an improved PID-fused adaptive neuro-fuzzy inference system (ANFIS); herein, ANFIS replaces composition controllers, improving the real-time performance of the control system. The ANFIS algorithm, however, suffers from problems such as easily falling into a local optimum solution. Therefore, the genetic algorithm (GA) and particle swarm optimisation (PSO) algorithm are separately used to optimise the network parameters of ANFIS, and the impact of different number of clusters (NC) on the performance of the control scheme is evaluated. The control scheme of PSO-optimised PID-ANFIS exhibits superior dynamic performance compared to other control schemes; moreover, it does not require the use of component controllers and has a faster response time and smaller transient values.