Venema, R. S.Bron, J.Zijlstra, R.M.Nijhuis, J. A. G.Spaanenburg, L.Haasis, H.-D.Ranze, K.C.2019-09-162019-09-161998https://dl.gi.de/handle/20.500.12116/26510One of the main issues in the research into a time series is its prediction. Artificial neural networks are suitable for that purpose because of their ability to identify non-linear systems. We illustrate the use of neural networks by a forecasting problem in waste-water purification, namely the prediction of its ammonia concentration. For this application, we used a feedforward architecture with an input delay line. However, because of the multi-variate, multi-scale and multi-stationary properties of the problem, we propose to put modularity in the neural design to capture these dynamics.Text/Conference Paper