Neural network-based software sensors for the estimation of key components in brewery wastewater anaerobic digester: an experimental validation

Abstract

Abstract This work focused on the experimental validation of software sensors with a view to improving on-line anaerobic digester monitoring. Based on cheaply available measurements such as conductivity, temperature, pH, redox potential, total suspended solids concentration and digester inflows and outflows, an intelligent estimator was built to reproduce the evolutions of key components such as volatile fatty acid, carbonate and alkalinity concentrations, as well as biogas composition (methane and carbon dioxide). The proposed solution considers a principal component pre-processing of the data selected as inputs of a radial basis function neural network (RBF-ANN) structure, using a particular sequential learning algorithm. Process dynamics were also taken into account, introducing a moving horizon version of this network (MH-RBF-ANN). Experimental results demonstrated the capacity of the MH-RBF-ANN to correctly predict the key-component evolutions and to improve the estimation accuracy, compared to the classical RBF-ANN.