Wind energy is a crucial component of the global transition to sustainable energy sources. The optimisation of wind farms can enhance their efficiency and output. This Innovation Study aims to significantly advance the real-time simulation and control of wind farms. Key to achieving this is the development of matrix-free optimisation techniques, which allow for the management of numerous constraints, and the implementation of parallelization across both spatial and temporal dimensions to fully exploit the computational power of large supercomputers.
Currently, numerical libraries are not readily accessible due to their incompatibility with the complex and efficiently parallelised data structures used by SP-Wind. This study addresses these limitations in order to achieve a potential order-of-magnitude speed-up in the Large Eddy Simulation (LES)-based optimal control of wind farms, thereby enabling real-time application using the following key methods:
- The central focus of this study is matrix-free optimization, which allows for the incorporation of numerous constraints without the computational overhead associated with traditional matrix-based methods. This approach is essential for handling the high-dimensional, nonlinear partial differential equations (PDEs) that describe wind farm dynamics.
- This study employs parallelization at both spatial and temporal levels. This dual-level parallelization is designed to maximize the use of large-scale supercomputing resources, thereby accelerating simulation times and enabling real-time control capabilities.
In order to achieve the project objectives, a matrix-free parallel multiple shooting method will be developed for the solution of large-scale nonlinear partial differential equations. This method will be integrated with advanced optimization libraries and will feature a user-friendly software interface, which will facilitate the optimization, time-parallelization, and simulation components. The framework will be implemented and validated using SP-Wind, with a particular focus on wind farm control.
The techniques and software developed in this study have broader implications beyond wind farm optimisation. The parallelisation strategies may be applied to any PDE problem that employs a time-marching approach. Potential applications include, but are not limited to, turbulence-resolving simulations in various fields, magneto-hydrodynamics, and finite-element methods for structural or electromagnetic problems formulated in the time domain.
The objective is to overcome current limitations in numerical libraries and to exploit the full potential of large supercomputers through advanced parallelisation and optimisation techniques, with the aim of delivering a transformative impact on the efficiency and effectiveness of wind energy systems. The anticipated advancements will not only benefit wind energy but also a multitude of other applications that necessitate high-performance computational solutions.