Wastewater Treatment Plants (WWTPs) control and prediction under a wide
range of operating conditions is an important goal in order to avoid
breaking of environmental balance, keep the system in stable operating
conditions and suitable decision-making. In this respect, the
availability of models characterizing WWTP behaviour as a dynamic
system is a necessary first step. However, due to the high complexity
of the WWTP processes and the heterogeneity, incompleteness and
imprecision of WWTP data, finding suitable models poses substantial
problems. In this work, an approach via soft computing techniques is
sought, in particular, by experimenting with fuzzy heterogeneous
time-delay neural networks to characterize the time variation of
outgoing variables. Experimental results show that these networks
are able to characterize WWTP behaviour in a statistically satisfactory
sense and also that they perform better than other well-established
neural network models.