Abstract
This study reports the development of an artificial neural network (ANN) representation of a bench scale in-line pH process. The dynamics of the process are characterised by a strong nonlinearity familiar to pH and are further complicated by variable flow which constantly changes the process deadtime. The focus of this paper is on determining a parsimonious ANN model structure which accommodates the variable process deadtime. Three approaches are compared; the first method, which does not require flow data, is to use a delay spread to include all possible values of the delayed variable in the model structure. The other two techniques make use of flowrate; one uses a volume array to represent the deadtime while the other samples at constant intervals of volume thus making the delay a constant number of samples. Results show that the three approaches lead to ANN models with similar prediction accuracy but differ in their computational complexity.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have