Author(s)

J. N. Fullerton, G. A. Walters & D. A. Savic

Abstract

A major challenge in the UK water industry is to improve the function and condition of sewer systems. Most of these are old, and convey both storm water and foul (sanitary) flows. Due to limited treatment capacities, overflows are provided to allow exceptional storm flows to spill into natural water courses. Unfortunately, many systems cause pollution by spilling excessively due to under-capacity, poor condition and poor maintenance. The water industry is investing large sums of money to rectify these problems. At present, the tools available are generally simulation packages such as HydroWorks, which produce representative storm flows to test existing and proposed sewer systems under design conditions. Designed to model network flows as accurately as possible, the time taken for a fill network simulation tends to be quite long. Although satisfactory for the traditional design process, any non-trivial system will have an enormous number of possible combinations of design variables. A typical problem which adds just 10 new network elements (additional pipes, increased storage, etc.) in, say, 20 locations would give a problem size of 1020. To search a problem of this size, Genetic Algorithms provide the best way (Walters [1], Savic [2]) of finding the global optimum. However, such a search will require tens of thousands of simulations of potential solutions to be carried out. Use of HydroWorks or similar commercial software would be prohibitively time-consuming. This paper describes work on formulating a very simple sewer flow model based on continuity, storage and times of flow, comparison of that model to HydroWorks, and the results of incorporating the model into a Genetic Algorithm (GA) optimiser. In particular, the results of comparing the two simulators on the optimised design are discussed in detail.

Keywords