Full metadata record
|dc.description.abstract||The mathematical modeling has played a leading role in the chemical process design and optimization. However, it is not straightforward to fully materialize the process modeling with mathematical formulation, in practice, because of complex design interactions existed. To remedy this problem, in this study, it is proposed to apply artificial neural network for chemical process design and optimization. Firstly, post-combustion CO2 capture membranes from coal-fired power plants was modelled based on artificial neural network structure. These surrogate models were directly employed in commercial simulator and the existing mathematical modeling fully substituted for this methodology. Secondly, most applications of reinforcement learning are focused on process control. In this work, it is suggested that this algorithm can be used for process optimization framework. Liquefaction process optimization was conducted with Deep Q-Network playing a role as learning optimal behavior policy and succeeding in acquiring optimal solution. UniSim Design®, MATLAB® and Python® are linked so that artificial neural networks can interact organically with commercial simulators in these works. Consequently, it was confirmed that the artificial neural network can be applied flexibly and sensitively by adjusting the structure of the hidden layer depending on problems of form and scale.||-|
|dc.title||Application of Machine Learning for Chemical Process Design and Optimization||-|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.