|
Research interests
In the last years, the research activity was focus on artificial intelligence tools (neural networks and evolutionary algorithms) used for modeling and optimization.
Methodologies based on different types of neural networks, including stacked neural networks and hybrid models which combine neural networks with phenomenological ones, have been developed.
Neural network were used in different types of applications: direct and inverse modeling, inferential modeling, soft sensors, optimal control.
In order to optimize different complex chemical processes, methods based on classical algorithms, biologically inspired and hybrid algorithms have been used. The optimization problems were formulated with scalar or vectorial objective functions.
A special optimization direction was represented by developing the optimal neural network topology. Methods based on evolutionary algorithms, used individually or combined with local search methods were used.
As optimizers, a series of biologically inspired algorithms, designed in different variants, have been applied: genetic algorithms, differential evolution, artificial immune systems (clonal selection), swarm intelligence algorithms (glow warm swarm optimizer, particle swarm optimization).
Different classification algorithms, including support vector machines, were also applied in chemical engineering field.
New software products being registered as patents were developed for neural network modeling and multi-objective optimization.
The modeling and optimization methodologies have been applied on various processes selected from chemical engineering field: polymerization processes, bio-processes, electrochemical processes, and other physico-chemical processes
Concerning the approached case studies, various types of problems were solved: estimating the reaction characteristics depending on the working conditions; predicting the final properties depending on the initial compound structures and/or reaction conditions; identifying the optimal working conditions that lead to pre-defined structures or features of the reaction; identifying the structures that will generate pre-established final properties (molecular design).
|
