Surrogate-based optimization for overflow spillway design

Overflow spillways are essential structures of dam projects, releasing surplus flood water to ensure dam safety. The discharge capacity of these structures depends mainly on their geometry; the most common shape is the standard weir, designed empirically. The design of overflow spillways must be optimised with respect to hydraulic performance, dam safety requirements and economic cost. The objective of this work is to numerically investigate spillway design using a shape optimisation process based on computational fluid dynamics software. The program Code_Saturne, which uses the volume of fluid method, was used to optimise the spillway design. To decrease computational cost, two reduced-order models were constructed. The first, with independent error formulation, was based on a neural network model, and the second was based on a kriging model, which considers correlated error formulation. The spillway optimisation was performed by a metaheuristic using each surrogate model as a simulator, and results are compared with those of the standard weir. The numerical optimisation results show an improvement over standard spillway hydraulic performances. Based on the optimised spillway shapes, expected gains are up to 15% for the flow discharge.