Analysis of uncertainties in water systems using neural networks

摘要

Modeling and simulation of all complex engineering systems is subject to a degree of uncertainty. Every effort should be made to ensure that the mathematical models reflect the physical system operations as accurately as possible. However, it isargued here that some uncertainty is unavoidable and the potential effects of this uncertainty must not be ignored.Recent research has resulted in a number of methods for assessing the influence of uncertainty in mathematical models or measurements. They are referred to in the literature as sensitivity analysis, error propagation, perturbation theory, or confidencelimit analysis. A characteristic feature of all these methods is that they are computationally demanding.This paper makes a contribution to the future development of real-time decision support in water distribution systems by demonstrating that the recurrent neural networks have potential for a dramatic improvement of computational efficiency of stateestimation and confidence limit analysis.It is envisaged that, with the current rate of development in the electronics industry and the emergence of new technologies, an implementation that takes the full advantage of the inherent parallelism of neural networks will be realized in the very nearfuture.