Modelling and Optimization of CO2 Absorption in Pneumatic Contactors Using Artificial Neural Networks Developed with Clonal Selection-Based Algorithm

Abstract

Abstract Our research focuses on the application of airlift contactors (ALRs) for the decontamination of CO2-containing gas streams, such as biogas. To assess the performance of ALRs during CO2 absorption, a complex experimental programme was applied in a laboratory-scale rectangular pneumatic contactor, able to operate either as a bubble column or as an airlift reactor. Using the experimental data, a model based on artificial neural network (ANN) was developed. The algorithm for determining the optimal neural network model and for reactor optimization is clonal selection (CS), belonging to artificial immune system class, which is a new computational intelligence paradigm based on the principles of the vertebrate immune system. To improve its capabilities and the probability for highly suitable models and input combinations, addressing maximum efficiency, a Back-Propagation (BK) algorithm – a supervised learning method based on the delta rule – is used as a local search procedure. It is applied in a greedy manner for the best antibody found in each generation. Since the highest affinity antibodies are cloned in the next generation, the effect of BK on the suitability of the individuals propagates into a large proportion of the population. In parallel with the BK hybridization of the basic CS–ANN combination, a series of normalization procedures are included for improving the overall results provided by the new algorithm called nCS-MBK (normalized Clonal Selection-Multilayer Perceptron Neural Network and Back-Propagation algorithm). The optimization allowed for achieving the optimal reactor configuration, which leads to a maximum amount of CO2 dissolved in water.