Building on Time-X results, the Inno4Scale project TiPOFlow will enable time-parallel optimization of turbulent flows in wind energy applications such as wind turbine farms.
Current HPC hardware enable optimal control and data assimilation of computational fluid dynamics with applications in wind energy and micrometeorology. However, the size of the PDE constrained optimization problems remains extremely challenging for real applications. Especially, real-time simulations and control are out of reach. The current in-house software simulation tool SP-Wind scales efficiently up to 200,000 compute cores. Recently, we started developing time-parallel multiple shooting methods in SP-Wind, showing good speed-ups in simple 2D and 3D tracking problems. However, generalization to turbulent boundary layers and windfarms remains challenging.
The goal of the project is to make an important step towards real time simulation and control of wind farms. Matrix-free optimization is required to enable many constraints, and parallelization both in space and time levels is used to fully exploit the capacity of large supercomputers. In addition, current state-of-the art numerical libraries are not easily accessible, as they do not work with the complex (and efficiently parallelized) data structures of SP-Wind. The current project aims at closing these gaps, leading to a potential order of magnitude speed-up of LES-based optimal control of wind farms, and enabling real-time use.
To achieve the goals, we will develop a matrix-free parallel multiple shooting method for large-scale nonlinear PDEs, efficiently coupled with optimization libraries, and develop an easy-to-use software interface between the optimization, time-parallelisation, and simulation software layers. The project relies on the complementary expertise of three research groups at KU Leuven, each renowned for their respective expertise. The proof-of-concept demonstrator in the current project is based on SP-Wind with a focus on wind-farm control. The project will also impact computational speed-up of other applications by parallelization of any PDE problem that relies on a time marching approach. Such applications are turbulence resolving simulations in various applications, magneto-hydrodynamics, but may also include finite-element methods for structural or electromagnetic problems when formulated in the time domain.