Model-free learning control of neutralization processes using reinforcement learning

Model-Free Learning Control of Chemical Processes

“…Since MFLC allows a tolerance error of the process whenever the pH is within the control band, the control signal is very smooth when the pH is closer or within the control band, even if some exploration is carried out. The detailed discussion of the application MFLC to pH control is given in Syafiie et al (2007a).…”

Adaptive nonlinear control of pH neutralization processes using fuzzy approximators

“…A fuzzy control algorithm combined with Takagi-Sugeno (T-S) fuzzy model (Pishvaie & Shahrokhi, 2006), a fuzzy gain scheduled control scheme (Regunath & Kadirkamanathan, 2001), a fuzzy self-tuning PI control (Babuska, Oosterhoff, Oudshoorn, & Bruijn, 2002), fuzzy sliding mode control (Shahraz & Boozarjomehry, 2009), fuzzy internal model control (Edgar & Postlethwaite, 2000;Han, Han, & Guo, 2006), neural networks and adaptive controller (Krishnapura & Jutan, 2000), PID controller using linearization through neural networks (Chen & Huang, 2004), a linear internal model controller optimized by a genetic algorithm (Mwembeshia, Kenta, & Salhi, 2004), a multiple model predictive controller based on T-S fuzzy model (He, Cai, & Li, 2005), iterative nonlinear model predictive control (Cueli & Bordons, 2008) and a model-free learning control based on reinforcement learning algorithms (Syafiiea, Tadeoa, & Martinez, 2007) have been reported for pH control. Some other works in this category can be found in the literature (Karr & Gentry, 1993;Loh, Looi, & Fong, 1995;Qin & Borders, 1994).…”

Model‐based reinforcement learning for nonlinear optimal control with practical asymptotic stability guarantees

“…Reinforcement learning (RL) is a popular algorithm for determining a policy that optimizes a user‐defined cost function through interactions with an environment 1‐7 . There are several applications of RL in chemical process engineering, such as a model‐free learning controller for pH control, 8 a policy gradient approach for batch bioprocess systems, 9 and approximate dynamic programming (ADP) for control of fed‐batch bioreactors, 10 control of proppant concentrations inside a fracture for hydraulic fracturing, 11 and control of alkali‐surfactant‐polymer flooding for oil recovery 12 . However, compared with the computer science field that can put aside the stability and mainly focus on optimality such as winning games or maximizing rewards, 13 RL applications in chemical processes have not become prevalent yet.…”