Optimisation algorithms could potentially provide extremely valuable guidance towards improved intervention strategies and/or designs for water systems. The application of these algorithms in this domain has historically been hindered by the extreme computational cost of performing hydraulic modelling of water systems. This is because running an optimisation algorithm generally involves running a very large number of simulations of the system being optimised. In this paper, a novel optimisation approach is described, based upon the ‘learning evolution model for multi-objective optimisation’ algorithm. This approach uses deep learning artificial neural network meta-models to reduce the number of simulations of the water system required, without reducing the accuracy of the optimisation results. This is then compared to an industry standard optimisation approach, showing results with increased speed of convergence and equivalent or improved accuracy. Therefore, demonstrating that this approach is suitable for use in highly computationally demanding areas such as water systems optimisation.