Predicting Transit Times and Flow Concentrations of Contaminants Through a Metropolitan Sewage System
by
J. Crawford
ANSTO
Coauthors: Barry J M, Airey P (ANSTO)

The interconnections of a typical metropolitan sewage system are highly complex and there is great uncertainty and variation in the myriad of input loads placed on the facility. Conventional mathematical modelling is not a simple task. ANSTO is a nuclear organisation which regularly makes controlled and authorised discharges into the system. It is useful to have a tool which can predict both the transit time through the system and the  concentration of trace amounts of radioactive liquid in the effluent as it passes through the treatment plant before discharge into the ocean.

Due to the complexity of the system, it was decided that this project  may well be modelled with a neural network, with the aim of using the network as a tool to predict future behaviour in the system. In addition, it was considered this was a suitable topic to evaluate the potential use of neural networks in an Environmental Dynamics research project being conducted.

A neural network system was developed and trained on data recorded for several routine discharges from ANSTO.  The neural network is compared with a more  empirical model of the system obtained through the use of piecewise linear approximation functions.  The two approaches are evaluated and compared.  

Date received: June 24, 1999