Neural networks and differential evolution algorithm applied for modelling the depollution process of some gaseous streams

• Published in: Environmental science and pollution research international Vol. 21; no. 22; pp. 12856 - 12867
• Main Authors: Curteanu, Silvia, Suditu, Gabriel Dan, Buburuzan, Adela Marina, Dragoi, Elena Niculina
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer-Verlag 01.11.2014; Springer Berlin Heidelberg; Springer Nature B.V
• Subjects: Activated carbon; Adsorbents; Adsorption; Air Pollutants - isolation & purification; Algorithms; Aquatic Pollution; Atmospheric Protection/Air Quality Control/Air Pollution; Back propagation; Catalytic oxidation; Charcoal - chemistry; Chemical engineering; Computer Simulation; Criteria; Dynamic tests; Earth and Environmental Science; Ecotoxicology; Environment; Environmental Chemistry; Environmental Health; Environmental impact; Environmental protection; Environmental Restoration and Remediation; Fitness; Gases; Genetic algorithms; Hexanes - isolation & purification; Hydrocarbons; Kinetics; Mathematical models; Membrane separation; mutation; Mutations; n-Hexane; Nanostructured materials; Neural networks; Neural Networks (Computer); Principal components analysis; Research Article; Resins; Sensors; Streams; VOCs; Volatile organic compounds; Waste Water Technology; Water Management; Water Pollution Control; Backpropagation; Differential evolution; Local search; Neural network; Hypercross-linked polymeric adsorbents; VOC adsorption
• ISSN: 0944-1344; 1614-7499; 1614-7499
• DOI: 10.1007/s11356-014-3232-x

The depollution of some gaseous streams containing n-hexane is studied by adsorption in a fixed bed column, under dynamic conditions, using granular activated carbon and two types of non-functionalized hypercross-linked polymeric resins. In order to model the process, a new neuro-evolutionary approach is proposed. It is a combination of a modified differential evolution (DE) with neural networks (NNs) and two local search algorithms, the global and local optimizers, working together to determine the optimal NN model. The main elements that characterize the applied variant of DE consist in using an opposition-based learning initialization, a simple self-adaptive procedure for the control parameters, and a modified mutation principle based on the fitness function as a criterion for reorganization. The results obtained prove that the proposed algorithm is able to determine a good model of the considered process, its performance being better than those of an available phenomenological model.