Concurrent Geometry, Control, and Layout Optimization of Wave Energy Converter Farms in Probabilistic Irregular Waves using Surrogate Modeling (2407.07098v1)

Abstract: A promising direction towards improving the performance of wave energy converter (WEC) farms is to leverage a system-level integrated approach known as control co-design (CCD). A WEC farm CCD problem may entail decision variables associated with the geometric attributes, control parameters, and layout of the farm. However, solving the resulting optimization problem, which requires the estimation of hydrodynamic coefficients through numerical methods such as multiple scattering (MS), is computationally prohibitive. To mitigate this computational bottleneck, we construct data-driven surrogate models (SMs) using artificial neural networks in combination with concepts from many-body expansion. The resulting SMs, developed using an active learning strategy known as query by committee, are validated through a variety of methods to ensure acceptable performance in estimating the hydrodynamic coefficients, (energy-related) objective function, and decision variables. To rectify inherent errors in SMs, a hybrid optimization strategy is devised. It involves solving an optimization problem with a genetic algorithm and SMs to generate a starting point that will be used with a gradient-based optimizer and MS. The effectiveness of the proposed approach is demonstrated by solving a series of optimization problems with increasing levels of integration. For a layout optimization study, the framework offers a 91-fold increase in computational efficiency compared to MS. Previously unexplored investigations of much further complexity are also performed, leading to a concurrent geometry, control, and layout optimization of WEC devices in probabilistic irregular waves. The scalability of the method is evaluated by increasing the farm size to include 25 devices. The results indicate promising directions toward a practical framework for integrated WEC farm design with more tractable computational demands.