Model-Free Learning Control of Chemical Processes

Syafiie, S.; Tadeo, Fernando; Martínez, Ernesto

doi:10.5772/5287

Cited by 5 publications

(5 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Both approaches would be theoretically appropriate to fulfill the control task. Regarding the current literature, most of the machine learning‐based control papers refer to model‐free reinforcement learning approaches in the domain of process control 5–8. It is worth mentioning that the first paper proposing the use of (deep) reinforcement learning for process control came already up in 1992.…”

Section: Process Control Using Reinforcement Learningmentioning

confidence: 99%

“…Bringing machine learning based controllers into the domain of process control is already a broad field of research for decades 5–9. However, most of the approaches come with either the disadvantage that the cost function of the LMPC optimization problem cannot be changed after the training process or the resulting controller requires a simulation environment for pre‐training and data sampling.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Investigation of a Model‐Based Deep Reinforcement Learning Controller Applied to an Air Separation Unit in a Production Environment

Blum

Krespach

Zapp

et al. 2021

Chemie Ingenieur Technik

View full text Add to dashboard Cite

The need for load flexibility and increased efficiency of energy-intensive processes has become more and more important in recent years. Control of the process variables plays a decisive role in maximizing the efficiency of a plant. The widely used control models of linear model predictive controllers (LMPC) are only partly suitable for nonlinear processes. One possibility for improvement is machine learning. In this work, one approach for a purely data-driven controller based on reinforcement learning is explored at an air separation plant (ASU) in productive use. The approach combines the model predictive controller with a data-generated nonlinear control model. The resulting controller and its control performance are examined in more detail on an ASU in real operation and compared with the previous LMPC solution. During the tests, stable behavior of the new control concept could be observed for several weeks in productive operation.

show abstract

Section: Process Control Using Reinforcement Learningmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Investigation of a Model‐Based Deep Reinforcement Learning Controller Applied to an Air Separation Unit in a Production Environment

Blum

Krespach

Zapp

et al. 2021

Chemie Ingenieur Technik

View full text Add to dashboard Cite

show abstract

“…Deep Reinforcement learning (RL) is emerging area of research in ML whereby an ANN is trained to take optimal actions to maximize a reward (or minimize a penalty) through continuous feedback during training [21]. RL has found remarkable success in a range of applications including complex gameplay, robotic control, autonomous navigation, and chemical process control, among others [22][23][24][25][26][27][28][29]. A key practice in RL is to use simulated data to train an RL agent such that it is capable of performing the same task in a real environment, otherwise known as sim-to-real transfer learning.…”

Section: Introductionmentioning

confidence: 99%

“…In deep RL, an artificial neural network (ANN) is trained to take optimal actions to maximize a reward (or minimize a penalty) through continuous feedback during training [25]. RL has found remarkable success in a range of applications including complex gameplay, robotic control, and autonomous navigation [26][27][28][29][30][31][32][33]. The goal of this paper is to present a proof-ofconcept demonstration of deep RL for controlling the thermal dynamics of APPJ-treated substrates with drastically different electrical and thermal properties.…”

Section: Introductionmentioning

confidence: 99%

Sim-to-real transfer reinforcement learning for control of thermal effects of an atmospheric pressure plasma jet

Witman

Gidon

Graves

et al. 2019

Plasma Sources Sci. Technol.

View full text Add to dashboard Cite

Applications of atmospheric pressure plasma jets (APPJs) present challenging controls problems due to the complexity of the plasma/substrate interactions. APPJs are particularly sensitive to changes in the electrical and thermal properties of treated substrates exhibiting large variations in thermal dynamics. The increasingly popular reinforcement learning (RL) based methods have great potential to aid regulation of thermal properties of APPJs across di↵erent substrates. Using only simulated data generated by a simple lumped-parameter energy balance, we train a robust reinforcement controller to perform temperature setpoint tracking by exposing it to randomized dynamics that capture the diverse temperature responses of di↵erent substrates. This approach is an e cient, safe, and flexible means for training an e↵ective RL controller which, when transferred to the live environment, is able to perform e↵ective temperature control over a wide variety of substrates without further fine-tuning.

show abstract

“…For example, in our previous studies on autonomous robot systems such as an intelligent wheelchair, we used RL algorithms for an agent in order to learn how to avoid obstacles and evolve cooperative behavior with other robots (Hamagami & Hirata, 2004;. Furthermore, RL has been widely used to solve the elevator dispatching problem (Crites & Barto, 1996), air-conditioning management problem (Dalamagkidisa et al, 2007), process control problem (S. Syafiie et al, 2008), etc. However, in most cases, RL algorithms have been successfully used only in ideal situations that are based on Markov decision processes (MDPs).…”

Section: Introductionmentioning

confidence: 99%